Update types and imports, add train script and utils and example input files

README.md

@@ -52,7 +52,7 @@ We provide several command line interfaces for quick usage of MuLAN different ap
 - `mulan-predict` for $\Delta \Delta G$ prediction of single and multiple-point mutations;
 - `mulan-att` to extract residues weights, related to interface regions;
 - `mulan-landscape` to produce a full mutational landscape for a given complex;
-- `mulan-train` to re-train the model on a custom dataset or to run cross validation (not supported yet);
+- `mulan-train` to re-train the model on a custom dataset or to run cross validation;
 - `plm-embed` to extract embeddings from protein language models. 
 Since the script uses the `transformers` interface, only models that are saved on HuggingFace 🤗 Hub can be loaded.
 

examples/S1102.tsv

+1A22_A	1A22_B	SA51A	0.3480601775656975
+1A22_A	1A22_B	FA165A	0.4104498198934685
+1A22_A	1A22_B	SA62A	0.1553600275920779
+1A22_A	1A22_B	IA168A	0.8059076005509755
+1A22_A	1A22_B	QA22A	-0.2197268133916914
+1A22_A	1A22_B	NA63A	0.3142135987668766
+1A22_A	1A22_B	RA172A	0.5425851484586328
+1A22_A	1A22_B	FA180A	0.1911951368906681
+1A22_A	1A22_B	EA56A	0.4104498198934685
+1A22_A	1A22_B	TA164A	1.906069936704197
+1A22_A	1A22_B	RA167A	2.4244821810979293
+1A22_A	1A22_B	HA18A	-0.4861468519007541
+1A22_A	1A22_B	QA68A	0.5881579306311284
+1A22_A	1A22_B	PA48A	0.4104498198934685
+1A22_A	1A22_B	KA161A	2.014032361204464
+1A22_A	1A22_B	LA45A	1.2239008844916306
+1A22_A	1A22_B	EA65A	-0.472839462221236
+1A22_A	1A22_B	DA160A	0.7905261144613256
+1A22_A	1A22_B	CA171A	1.0098644391267992
+1A22_A	1A22_B	HA21A	0.1553600275920779
+1A22_A	1A22_B	KA157A	-0.1547681695521046
+1A22_A	1A22_B	EA163A	-0.9241436813966502
+1A22_A	1A22_B	DA26A	-0.2112064661099832
+1A22_A	1A22_B	YA153A	0.3480601775656975
+1A22_A	1A22_B	QA46A	0.1079624245002666
+1A22_A	1A22_B	PA61A	1.20871661869346
+1A22_A	1A22_B	RA64A	1.641799279573867
+1A22_A	1A22_B	RA156A	0.2783144913283042
+1A22_A	1A22_B	YA42A	0.1992433537834852
+1A22_A	1A22_B	RB11M	0.999896442915034
+1A22_A	1A22_B	SB174A	0.0338465787988226
+1A22_A	1A22_B	RB11L	0.5213078683529346
+1A22_A	1A22_B	SB58A	-0.5066860410200569
+1A22_A	1A22_B	CB81A	0.0
+1A22_A	1A22_B	WB63F	2.783407873049008
+1A22_A	1A22_B	CB67A	0.0
+1A4Y_A	1A4Y_B	WA261A	0.1006063805906443
+1A4Y_A	1A4Y_B	WA375A	1.0348412327905765
+1A4Y_A	1A4Y_B	EA344A	0.178519836594301
+1A4Y_A	1A4Y_B	YA434F	0.5575771876945907
+1A4Y_A	1A4Y_B	WA263A	1.170784638271947
+1A4Y_A	1A4Y_B	YA434A	3.2603738601208216
+1A4Y_A	1A4Y_B	DA435A	3.4837901179692
+1A4Y_A	1A4Y_B	WA318A	1.4999902762747832
+1A4Y_A	1A4Y_B	RA457A	-0.2238733431043407
+1A4Y_A	1A4Y_B	YA437A	0.8358916790228896
+1A4Y_A	1A4Y_B	SA289A	0.0423147056193045
+1A4Y_A	1A4Y_B	EA401A	0.8832892821147027
+1A4Y_A	1A4Y_B	KA320A	-0.3098434393028242
+1A4Y_A	1A4Y_B	IA459A	0.6790906364878461
+1A4Y_A	1A4Y_B	YA437F	0.2451373913115802
+1A4Y_A	1A4Y_B	HB84A	0.1703520668280376
+1A4Y_A	1A4Y_B	KB40G	3.2351720176220624
+1A4Y_A	1A4Y_B	RB5A	2.3081202693877465
+1A4Y_A	1A4Y_B	HB13A	-0.296536049623306
+1A4Y_A	1A4Y_B	EB108A	-0.3225623321016648
+1A4Y_A	1A4Y_B	RB31A	0.2503963416825741
+1A4Y_A	1A4Y_B	WB89A	0.2400977530654273
+1A4Y_A	1A4Y_B	KB40Q	4.250288323419703
+1A4Y_A	1A4Y_B	RB32A	0.9097928484168456
+1A4Y_A	1A4Y_B	QB12A	0.3001140298124305
+1A4Y_A	1A4Y_B	HB114A	0.6564989187287935
+1A4Y_A	1A4Y_B	NB68A	0.1177503154404853
+1A4Y_A	1A4Y_B	HB8A	0.9036602217884386
+1ACB_A	1ACB_B	LB38S	4.950962020246454
+1ACB_A	1ACB_B	LB38P	6.852342536090567
+1ACB_A	1ACB_B	LB38G	6.0329104263395
+1ACB_A	1ACB_B	LB38E	6.553915374192577
+1ACB_A	1ACB_B	LB38I	4.23043510099434
+1ACB_A	1ACB_B	LB38D	6.771166322627187
+1AK4_A	1AK4_B	PB93A	2.0460484782066217
+1AK4_A	1AK4_B	HB87R	3.0083601989953013
+1AK4_A	1AK4_B	HB87A	2.37295805315554
+1AK4_A	1AK4_B	AB88V	2.13286030009011
+1AK4_A	1AK4_B	PB85A	2.4486550851628266
+1AK4_A	1AK4_B	AB92V	1.7224104801966431
+1AK4_A	1AK4_B	IB91A	1.6035825413109936
+1AK4_A	1AK4_B	AB88G	4.016996102578333
+1AK4_A	1AK4_B	GB89V	4.392941760129896
+1AK4_A	1AK4_B	AB92G	1.9442866749670664
+1AK4_A	1AK4_B	VB86A	2.3547364948605334
+1AK4_A	1AK4_B	HB87Q	2.3359363787083702
+1AK4_A	1AK4_B	PB90A	3.5354023574986417
+1AK4_A	1AK4_B	IB91V	1.3634847882455636
+1AK4_A	1AK4_B	GB89A	3.439906791777797
+1B2S_A	1B2S_B	AA27K	-3.864350158924305
+1B2S_A	1B2S_B	AB43T	-4.166151396743981
+1B2U_A	1B2U_B	AA27K	-5.043489333088793
+1B2U_A	1B2U_B	AB36D	-4.166151396743981
+1B3S_A	1B3S_B	AA102H	-5.679044866811731
+1B3S_A	1B3S_B	FB30Y	0.5957939259574623
+1BRS_A	1BRS_B	EA58A	0.0944460365287085
+1BRS_A	1BRS_B	EA71C	2.527489313754817
+1BRS_A	1BRS_B	HA100G	6.817423941227826
+1BRS_A	1BRS_B	RA57K	2.4868718234257017
+1BRS_A	1BRS_B	HA100D	4.548301787721975
+1BRS_A	1BRS_B	EA71W	1.6564211069067412
+1BRS_A	1BRS_B	HA100L	7.662496485079659
+1BRS_A	1BRS_B	RA81Q	5.4172993901379405
+1BRS_A	1BRS_B	EA71F	2.2340551775614266
+1BRS_A	1BRS_B	WA33F	1.2596283269819324
+1BRS_A	1BRS_B	EA71Y	2.412519068771428
+1BRS_A	1BRS_B	EA71S	3.0089735122761176
+1BRS_A	1BRS_B	HA100Q	4.548301787721975
+1BRS_A	1BRS_B	EA71Q	1.4513722760373606
+1BRS_A	1BRS_B	NA56A	3.089507324075856
+1BRS_A	1BRS_B	YB29A	3.4687980787332435
+1BRS_A	1BRS_B	TB42A	1.8566951901405384
+1BRS_A	1BRS_B	WB38F	1.641799279573867
+1BRS_A	1BRS_B	WB44F	0.0564382965578786
+1BRS_A	1BRS_B	YB29F	-0.132135328565166
+1CSE_A	1CSE_B	LB38D	4.350433681449669
+1CSE_A	1CSE_B	LB38G	2.2488334837936126
+1CSE_A	1CSE_B	LB38I	2.9409875695104866
+1CSE_A	1CSE_B	LB38E	2.349579985314381
+1CSE_A	1CSE_B	LB38S	1.1722287392442985
+1CSE_A	1CSE_B	LB38P	6.67127645335119
+1CSO_A	1CSO_B	IB13L	-4.4188804360352485
+1CT0_A	1CT0_B	SB13L	-4.100355769895383
+1CT2_A	1CT2_B	TB13L	-3.1601132172258635
+1CT4_A	1CT4_B	VB13L	-2.997860793780033
+1E96_A	1E96_B	NA26H	1.0854474195615502
+1E96_A	1E96_B	IA33N	2.0077513733569488
+1EAW_A	1EAW_B	TA144A	0.0894056741917026
+1EAW_A	1EAW_B	DA214A	2.2280123870440565
+1EAW_A	1EAW_B	FA92A	0.8915712901218367
+1EAW_A	1EAW_B	YA141A	0.5019009328018527
+1EAW_A	1EAW_B	FA89A	0.7282270437769274
+1EAW_A	1EAW_B	DA91A	0.6541111980754817
+1EAW_A	1EAW_B	RA48A	0.5873189040934879
+1EAW_A	1EAW_B	IA45A	-0.1942602619391777
+1EAW_A	1EAW_B	FA50A	-0.4285969168341221
+1EAW_A	1EAW_B	YA52A	-0.0789009539196516
+1EAW_A	1EAW_B	QA169A	-0.1330295959181544
+1EAW_A	1EAW_B	IA26A	-0.8220922966245396
+1EAW_A	1EAW_B	DA47A	1.5020192808795692
+1EAW_A	1EAW_B	HA138A	-0.0144729139862604
+1EAW_A	1EAW_B	QA23A	-0.5186505836683111
+1EFN_A	1EFN_B	TA13H	1.2429228688538476
+1EFN_A	1EFN_B	IA12A	1.4502966101290538
+1EMV_A	1EMV_B	IA51A	0.8476985078878343
+1EMV_A	1EMV_B	TA36A	0.8995136920757965
+1EMV_A	1EMV_B	SA26A	0.1730871937216989
+1EMV_A	1EMV_B	TA25A	0.7280937188813291
+1EMV_A	1EMV_B	VA35A	1.6637749495047736
+1EMV_A	1EMV_B	SA46A	0.0072657834232536
+1EMV_A	1EMV_B	EA28A	1.4159220295835304
+1EMV_A	1EMV_B	VA32A	2.5781527527089203
+1EMV_A	1EMV_B	NA22A	0.1394011119884197
+1EMV_A	1EMV_B	DA49A	5.915151095926328
+1EMV_A	1EMV_B	SA27A	0.9559519886336716
+1EMV_A	1EMV_B	YA53A	4.634499819532684
+1EMV_A	1EMV_B	EA39A	2.0829652073420992
+1EMV_A	1EMV_B	SA48A	2.18711955492616
+1EMV_A	1EMV_B	YA52A	4.834367918146082
+1EMV_A	1EMV_B	HA44A	0.8317651541723166
+1EMV_A	1EMV_B	CA21A	0.9215060203775778
+1EMV_A	1EMV_B	PA54A	1.242214974065785
+1EMV_A	1EMV_B	PA45A	0.4372486879943693
+1EMV_A	1EMV_B	LA31A	3.417558708632216
+1EMV_A	1EMV_B	GA47A	1.4852958766126645
+1EMV_A	1EMV_B	NB75A	2.3344109640223465
+1EMV_A	1EMV_B	QB92A	-0.2775883674853894
+1EMV_A	1EMV_B	SB78A	-0.5398667630280478
+1EMV_A	1EMV_B	FB86A	3.8787286930522846
+1EMV_A	1EMV_B	SB77A	-0.2328762743011872
+1EMV_A	1EMV_B	KB97A	1.959528419718087
+1EMV_A	1EMV_B	TB87A	0.1584260231997625
+1EMV_A	1EMV_B	VB98A	1.0887920447193764
+1EMV_A	1EMV_B	RB54A	1.6656087875428582
+1EMV_A	1EMV_B	SB84A	-0.1094173488678116
+1EMV_A	1EMV_B	NB72A	1.1646046076839518
+1EMV_A	1EMV_B	SB74A	-0.2410066617929089
+1F47_A	1F47_B	KA14A	-0.0426193899170961
+1F47_A	1F47_B	YA5A	0.8689295482454922
+1F47_A	1F47_B	DA7G	1.1387685212697871
+1F47_A	1F47_B	FA11A	2.444252433508604
+1F47_A	1F47_B	DA7A	1.732786408841286
+1F47_A	1F47_B	IA8A	2.514979319506033
+1F47_A	1F47_B	LA12A	2.2941716734520803
+1F47_A	1F47_B	DA7S	2.063822508640781
+1F47_A	1F47_B	DA4A	0.6911582373455776
+1F47_A	1F47_B	PA9A	-0.0575358944989963
+1F47_A	1F47_B	QA15A	-0.0455727821724947
+1F47_A	1F47_B	LA6A	0.9249655467883962
+1FC2_A	1FC2_B	YA10W	0.4104498198934649
+1FCC_A	1FCC_B	EB42A	0.3852479773947088
+1FCC_A	1FCC_B	WB43A	3.771285474126445
+1FCC_A	1FCC_B	DB40A	0.2721138700675905
+1FCC_A	1FCC_B	KB31A	3.4758058992163714
+1FCC_A	1FCC_B	TB25A	0.2400977530654309
+1FCC_A	1FCC_B	KB28A	1.2555223637452997
+1FCC_A	1FCC_B	NB35A	2.3639173596894745
+1FFW_A	1FFW_B	YA105W	0.7129372152866669
+1FFW_A	1FFW_B	AA89V	0.0912809292832159
+1FFW_A	1FFW_B	DB49A	0.0962362211265936
+1FFW_A	1FFW_B	DB44A	-0.0741158457014421
+1FFW_A	1FFW_B	CB55A	0.2041986456268584
+1FFW_A	1FFW_B	EB20A	0.6388213695852247
+1FFW_A	1FFW_B	HB23A	0.0338465787988226
+1FFW_A	1FFW_B	EB13A	0.7164102687598799
+1FFW_A	1FFW_B	IB58A	0.4276149034752397
+1FFW_A	1FFW_B	FB56A	3.644105437404656
+1FR2_A	1FR2_B	AA39E	-2.0829652073420992
+1GC1_A	1GC1_B	QB33A	0.1052146055570144
+1GC1_A	1GC1_B	KB46A	1.4302990134480849
+1GC1_A	1GC1_B	SB60A	-0.0885458023895573
+1GC1_A	1GC1_B	KB35A	0.3219040175039112
+1GC1_A	1GC1_B	QB40A	-0.4104498198934649
+1GC1_A	1GC1_B	NB32A	0.1825618585664354
+1GC1_A	1GC1_B	TB45A	-0.1488168237822105
+1GC1_A	1GC1_B	SB42A	0.0
+1GC1_A	1GC1_B	RB59A	1.1753222384043518
+1GC1_A	1GC1_B	LB44A	1.055488944574842
+1GC1_A	1GC1_B	QB25A	0.0320161170021613
+1GC1_A	1GC1_B	KB29A	0.5359688347135325
+1GC1_A	1GC1_B	DB63A	-0.3191688906102499
+1GC1_A	1GC1_B	NB52A	0.7079819234432936
+1GC1_A	1GC1_B	EB85A	1.3226303889636168
+1GC1_A	1GC1_B	HB27A	0.2824124586847354
+1GC1_A	1GC1_B	QB64A	0.4424659368956298
+1GC1_A	1GC1_B	SB23A	0.292534994403395
+1GL0_A	1GL0_B	MB28K	-0.5693033910682619
+1GL1_A	1GL1_B	LB29V	4.257093855643765
+1H9D_A	1H9D_B	VB3A	1.284109116566453
+1H9D_A	1H9D_B	LB93A	0.8608913724081031
+1H9D_A	1H9D_B	RB2A	1.0641537924494653
+1H9D_A	1H9D_B	QB66A	1.2490984805068432
+1H9D_A	1H9D_B	GB60A	1.9022255267258377
+1H9D_A	1H9D_B	NB94A	2.1099898529149463
+1HE8_A	1HE8_B	KA80V	0.4671409205936161
+1IAR_A	1IAR_B	RA53Q	0.8353431343614393
+1IAR_A	1IAR_B	QA8A	-0.0223480831455837
+1IAR_A	1IAR_B	FA82D	-0.5803147176938648
+1IAR_A	1IAR_B	WA91D	1.3059488937465673
+1IAR_A	1IAR_B	QA78A	0.124779284655542
+1IAR_A	1IAR_B	TA6D	1.3920278517745128
+1IAR_A	1IAR_B	RA85E	1.223993415770776
+1IAR_A	1IAR_B	RA88A	3.7528288874815807
+1IAR_A	1IAR_B	WA91A	0.7291616252477553
+1IAR_A	1IAR_B	QA78E	0.2449515008942047
+1IAR_A	1IAR_B	IA5R	0.7956977972881756
+1IAR_A	1IAR_B	KA84A	0.3448021010125046
+1IAR_A	1IAR_B	KA12E	0.1394011119884162
+1IAR_A	1IAR_B	IA5A	1.1707846382719451
+1IAR_A	1IAR_B	KA84D	1.8791492136960457
+1IAR_A	1IAR_B	QA8R	0.0389017052793203
+1IAR_A	1IAR_B	TA13A	0.9780844882983252
+1IAR_A	1IAR_B	FA82A	-0.0864271814544448
+1IAR_A	1IAR_B	KA12S	-0.0148046190983937
+1IAR_A	1IAR_B	RA85A	0.4266742298545552
+1IAR_A	1IAR_B	TA6A	-0.1035922650362177
+1IAR_A	1IAR_B	RA81A	0.4793826660310998
+1IAR_A	1IAR_B	RA81E	1.4609989742393452
+1IAR_A	1IAR_B	NA89A	1.5580097591384998
+1IAR_A	1IAR_B	TA13D	-0.2185625100196091
+1IAR_A	1IAR_B	EA9Q	3.111263236322809
+1IAR_A	1IAR_B	RA88Q	2.827575957081772
+1JCK_A	1JCK_B	YB90A	2.5948342479259647
+1JCK_A	1JCK_B	KB103A	0.6762974524423493
+1JCK_A	1JCK_B	FB176A	2.132860300090111
+1JCK_A	1JCK_B	YB26A	1.7739346081390313
+1JCK_A	1JCK_B	TB20A	1.6540090713122657
+1JCK_A	1JCK_B	VB91A	2.2317820311243093
+1JCK_A	1JCK_B	NB60A	1.641799279573867
+1JTG_A	1JTG_B	QA74A	0.4293857000092753
+1JTG_A	1JTG_B	VA78A	1.9097825126251613
+1JTG_A	1JTG_B	MA104A	0.7384493296917629
+1JTG_A	1JTG_B	VA190A	-0.4067372439725005
+1JTG_A	1JTG_B	GA212S	-1.6311295509134285
+1JTG_A	1JTG_B	EA79K	4.233259421962298
+1JTG_A	1JTG_B	SA209A	1.2384592701602788
+1JTG_A	1JTG_B	PA82A	-0.3825643399076011
+1JTG_A	1JTG_B	EA143A	-0.0725454735794386
+1JTG_A	1JTG_B	YA80A	-0.1682866516901349
+1JTG_A	1JTG_B	NA75A	-0.4555758620072812
+1JTG_A	1JTG_B	EA85A	4.059884794410223
+1JTG_A	1JTG_B	TB140K	-0.0143850131246843
+1JTG_A	1JTG_B	WB112A	3.0089966437403977
+1JTG_A	1JTG_B	YB143A	0.3818377849685177
+1JTG_A	1JTG_B	YB50A	-0.4067372439725005
+1JTG_A	1JTG_B	DB163A	-1.3402600892186527
+1JTG_A	1JTG_B	HB148A	2.746506323142203
+1JTG_A	1JTG_B	RB160A	2.2210862582653434
+1JTG_A	1JTG_B	DB163K	-1.9817669687114847
+1JTG_A	1JTG_B	SB113A	-0.1682866516901349
+1JTG_A	1JTG_B	WB150A	4.251421862409736
+1JTG_A	1JTG_B	SB71A	0.3579091184487169
+1JTG_A	1JTG_B	WB162A	2.3396489546293378
+1JTG_A	1JTG_B	YB53A	2.076422003532233
+1JTG_A	1JTG_B	FB36A	3.1998090387123685
+1JTG_A	1JTG_B	HB41A	3.2480654583719684
+1KAC_A	1KAC_B	SA87Y	-1.252128922253883
+1KAC_A	1KAC_B	PA15S	-0.7929814901412069
+1KTZ_A	1KTZ_B	RA64A	2.882953252292189
+1KTZ_A	1KTZ_B	RA64K	2.200389562331132
+1KTZ_A	1KTZ_B	RA13A	1.4808360332993633
+1KTZ_A	1KTZ_B	VA62I	0.242795497684547
+1KTZ_A	1KTZ_B	RA13K	1.1511183728524426
+1KTZ_A	1KTZ_B	IB29A	1.815898256424136
+1KTZ_A	1KTZ_B	TB27A	1.9591341277245409
+1KTZ_A	1KTZ_B	SB25A	0.7727305005342533
+1KTZ_A	1KTZ_B	EB95Q	2.0727094276112656
+1KTZ_A	1KTZ_B	DB94A	1.2606811735209271
+1KTZ_A	1KTZ_B	DB8N	2.445686888874384
+1KTZ_A	1KTZ_B	LB3A	2.2703545242309806
+1KTZ_A	1KTZ_B	VB53A	0.8611368905818617
+1KTZ_A	1KTZ_B	SB28A	0.6627573646972937
+1KTZ_A	1KTZ_B	SB28L	4.482104068477997
+1KTZ_A	1KTZ_B	EB31A	1.6622596077177991
+1KTZ_A	1KTZ_B	EB95A	1.939875392275132
+1KTZ_A	1KTZ_B	FB6A	3.424914752541837
+1KTZ_A	1KTZ_B	IB26A	2.3418757857708226
+1KTZ_A	1KTZ_B	DB8A	1.9671999130830256
+1LFD_A	1LFD_B	KA39A	1.178714348448343
+1LFD_A	1LFD_B	RA7A	1.1354722285130876
+1LFD_A	1LFD_B	KA19A	1.3256748888305028
+1LFD_A	1LFD_B	DA38K	-1.0859338739665807
+1LFD_A	1LFD_B	DA43A	-0.2795366876748657
+1LFD_A	1LFD_B	NA14K	0.4027693760390596
+1LFD_A	1LFD_B	DA38A	-0.5787704819172914
+1LFD_A	1LFD_B	MA17K	-0.9203766019163782
+1LFD_A	1LFD_B	NA41K	-1.1677401384050592
+1M9E_A	1M9E_B	AB76H	-2.37295805315554
+1MAH_A	1MAH_B	FA287Y	0.7919900197272476
+1MAH_A	1MAH_B	YA69N	5.2174270026893605
+1MAH_A	1MAH_B	FA287I	2.06036958390061
+1MAH_A	1MAH_B	FA285L	1.244947005000947
+1MAH_A	1MAH_B	DA71N	1.87946870323982
+1MAH_A	1MAH_B	YA121Q	3.004103816360712
+1MAH_A	1MAH_B	WA276R	8.808838044698076
+1P69_A	1P69_B	SA15P	0.0
+1P6A_A	1P6A_B	YA87S	0.0
+1PPF_A	1PPF_B	NB36H	-0.525587222200782
+1PPF_A	1PPF_B	AB15G	1.1223859378254897
+1PPF_A	1PPF_B	LB18V	-0.4858069628552393
+1PPF_A	1PPF_B	EB19Q	0.6512382651313118
+1PPF_A	1PPF_B	NB36I	0.8593760161521313
+1PPF_A	1PPF_B	KB13V	0.8593760161521313
+1PPF_A	1PPF_B	RB21P	6.612233374468008
+1PPF_A	1PPF_B	TB17P	3.031374527610504
+1PPF_A	1PPF_B	KB13G	1.231501458438531
+1PPF_A	1PPF_B	EB19C	1.3831093378131278
+1PPF_A	1PPF_B	LB18F	5.097540666949608
+1PPF_A	1PPF_B	GB32P	0.2601531651498519
+1PPF_A	1PPF_B	NB36M	1.0790916495695375
+1PPF_A	1PPF_B	NB36V	0.3410085695957665
+1PPF_A	1PPF_B	LB18H	6.381171448492668
+1PPF_A	1PPF_B	TB17Q	1.9413244222451755
+1PPF_A	1PPF_B	AB15M	-0.3957512809449515
+1PPF_A	1PPF_B	KB13R	-0.6319253388219295
+1PPF_A	1PPF_B	EB19Y	1.231501458438531
+1PPF_A	1PPF_B	TB17M	2.4675689168328567
+1PPF_A	1PPF_B	RB21C	-0.0800488939137569
+1PPF_A	1PPF_B	YB20N	3.593094703431716
+1PPF_A	1PPF_B	AB15T	0.9494316980624192
+1PPF_A	1PPF_B	NB36S	-1.3276379503571842
+1PPF_A	1PPF_B	AB15S	0.745948995933615
+1PPF_A	1PPF_B	YB20C	3.403623598200179
+1PPF_A	1PPF_B	EB19G	2.0911319749409856
+1PPF_A	1PPF_B	NB36W	1.7038071206990306
+1PPF_A	1PPF_B	LB18R	7.084539036728509
+1PPF_A	1PPF_B	GB32R	4.781497932955725
+1PPF_A	1PPF_B	LB18I	-0.7226819272889191
+1PPF_A	1PPF_B	GB32I	4.311461974288624
+1PPF_A	1PPF_B	PB14A	-0.1224011681871513
+1PPF_A	1PPF_B	NB36E	-1.004174409411604
+1PPF_A	1PPF_B	KB13D	0.6225010517184497
+1PPF_A	1PPF_B	EB19N	1.195813690056745
+1PPF_A	1PPF_B	PB14F	-1.8309899418626097
+1PPF_A	1PPF_B	KB13N	0.6512382651313118
+1PPF_A	1PPF_B	YB20P	5.29245463897787
+1PPF_A	1PPF_B	TB17G	3.549741974166869
+1PPF_A	1PPF_B	RB21H	-0.4858069628552393
+1PPF_A	1PPF_B	EB19D	0.5698805931962312
+1PPF_A	1PPF_B	AB15F	-0.071934797575297
+1PPF_A	1PPF_B	AB15N	0.902728745416976
+1PPF_A	1PPF_B	EB19A	1.186540391462982
+1PPF_A	1PPF_B	NB36C	0.5961414432378067
+1PPF_A	1PPF_B	EB19F	1.3451829790073706
+1PPF_A	1PPF_B	EB19K	2.0911319749409856
+1PPF_A	1PPF_B	YB20V	4.12763664026779
+1PPF_A	1PPF_B	YB20S	3.593094703431716
+1PPF_A	1PPF_B	KB13H	1.00043953246319
+1PPF_A	1PPF_B	RB21Q	-0.0180883038813046
+1PPF_A	1PPF_B	TB17D	4.855144693828697
+1PPF_A	1PPF_B	NB36F	1.8205139793267249
+1PPF_A	1PPF_B	YB20I	3.8769063494777214
+1PPF_A	1PPF_B	AB15R	-0.2605008030922313
+1PPF_A	1PPF_B	RB21V	-0.3447434465441788
+1PPF_A	1PPF_B	RB21A	0.2051566446635693
+1PPF_A	1PPF_B	GB32Y	1.186540391462982
+1PPF_A	1PPF_B	NB36P	-2.9184025459312224
+1PPF_A	1PPF_B	YB20G	4.12763664026779
+1PPF_A	1PPF_B	LB18P	6.061585379633843
+1PPF_A	1PPF_B	GB32M	2.1647787358139574
+1PPF_A	1PPF_B	GB32L	3.086332236383951
+1PPF_A	1PPF_B	KB13C	0.9494316980624192
+1PPF_A	1PPF_B	PB14V	-1.5038255665517577
+1PPF_A	1PPF_B	NB36Y	1.653500192854958
+1PPF_A	1PPF_B	GB32N	1.5914808178008322
+1PPF_A	1PPF_B	AB15P	1.9676840307258203
+1PPF_A	1PPF_B	YB20K	4.431515215391321
+1PPF_A	1PPF_B	GB32E	1.4372706822530503
+1PPF_A	1PPF_B	YB20T	4.984980635084527
+1PPF_A	1PPF_B	TB17W	2.1647787358139574
+1PPF_A	1PPF_B	KB13M	0.2051566446635693
+1PPF_A	1PPF_B	KB13W	0.8195957568065886
+1PPF_A	1PPF_B	KB13A	0.745948995933615
+1PPF_A	1PPF_B	RB21T	-0.0180883038813046
+1PPF_A	1PPF_B	KB13E	1.117146391090886
+1PPF_A	1PPF_B	KB13S	0.4544355898142811
+1PPF_A	1PPF_B	RB21F	-0.9675894583346132
+1PPF_A	1PPF_B	TB17C	2.058440619192808
+1PPF_A	1PPF_B	PB14L	-2.882675792827865
+1PPF_A	1PPF_B	KB13F	0.9494316980624192
+1PPF_A	1PPF_B	GB32D	2.462329370098253
+1PPF_A	1PPF_B	RB21E	0.4544355898142811
+1PPF_A	1PPF_B	NB36K	2.585275850295563
+1PPF_A	1PPF_B	RB21L	-0.8466354583335054
+1PPF_A	1PPF_B	EB19S	1.8205139793267249
+1PPF_A	1PPF_B	LB18Y	6.543423871938499
+1PPF_A	1PPF_B	RB21D	0.2051566446635693
+1PPF_A	1PPF_B	PB14D	-0.4858069628552393
+1PPF_A	1PPF_B	TB17Y	2.44632273745208
+1PPF_A	1PPF_B	PB14T	-0.2605008030922313
+1PPF_A	1PPF_B	EB19W	1.508778598563552
+1PPF_A	1PPF_B	RB21G	0.5698805931962312
+1PPF_A	1PPF_B	AB15Y	-0.1824910230983487
+1PPF_A	1PPF_B	AB15K	1.111859426999418
+1PPF_A	1PPF_B	GB32H	1.4840320759073862
+1PPF_A	1PPF_B	GB32W	1.621096016273885
+1PPF_A	1PPF_B	KB13Y	0.4985475206738652
+1PPF_A	1PPF_B	YB20R	4.311461974288624
+1PPF_A	1PPF_B	YB20W	0.232105913093271
+1PPF_A	1PPF_B	KB13P	1.2485597665171095
+1PPF_A	1PPF_B	PB14G	0.0920877032456797
+1PPF_A	1PPF_B	YB20M	2.773253383796048
+1PPF_A	1PPF_B	RB21K	0.5961414432378067
+1PPF_A	1PPF_B	YB20F	0.4753310003193558
+1PPF_A	1PPF_B	YB20A	3.165696958334093
+1PPF_A	1PPF_B	LB18C	-0.088277277051608
+1PPF_A	1PPF_B	LB18N	5.152695328112994
+1PPF_A	1PPF_B	KB13L	0.3959662783692117
+1PPF_A	1PPF_B	TB17N	2.661149593010588
+1PPF_A	1PPF_B	EB19V	1.1428864814382518
+1PPF_A	1PPF_B	AB15E	0.1041336051620867
+1PPF_A	1PPF_B	YB20Q	4.631690380259915
+1PPF_A	1PPF_B	NB36G	-0.5635830138822371
+1PPF_A	1PPF_B	PB14M	-1.5525375648393942
+1PPF_A	1PPF_B	TB17I	1.4726879224776823
+1PPF_A	1PPF_B	TB17L	2.2479117244243447
+1PPF_A	1PPF_B	PB14H	-1.7409342599523203
+1PPF_A	1PPF_B	EB19R	1.4493165841694555
+1PPF_A	1PPF_B	RB21Y	0.2175606253806012
+1PPF_A	1PPF_B	PB14I	-1.6283742280872389
+1PPF_A	1PPF_B	RB21M	-0.7490415357695621
+1PPF_A	1PPF_B	AB15W	-0.6319253388219295
+1PPF_A	1PPF_B	PB14K	-0.4424542335903929
+1PPF_A	1PPF_B	LB18S	3.0489900997063994
+1PPF_A	1PPF_B	GB32S	0.902728745416976
+1PPF_A	1PPF_B	KB13T	0.1635956195561831
+1PPF_A	1PPF_B	LB18D	7.507514860161978
+1PPF_A	1PPF_B	GB32Q	2.760756531996245
+1PPF_A	1PPF_B	AB15C	-0.664329515403006
+1PPF_A	1PPF_B	EB19H	0.6808534636043628
+1PPF_A	1PPF_B	PB14N	-0.6319253388219295
+1PPF_A	1PPF_B	AB15I	0.3248915324240951
+1PPF_A	1PPF_B	YB20E	6.330163614091897
+1PPF_A	1PPF_B	PB14Y	-0.9492167006381572
+1PPF_A	1PPF_B	NB36Q	0.3775935206727574
+1PPF_A	1PPF_B	AB15Q	0.2462978387934616
+1PPF_A	1PPF_B	AB15L	0.0175450286501863
+1PPF_A	1PPF_B	PB14Q	-0.7999908007261016
+1PPF_A	1PPF_B	TB17H	1.670074511431462
+1PPF_A	1PPF_B	PB14E	-1.4091148357494525
+1PPF_A	1PPF_B	TB17K	3.341604223146051
+1PPF_A	1PPF_B	AB15H	0.545192178281269
+1PPF_A	1PPF_B	RB21I	-0.7999908007261016
+1PPF_A	1PPF_B	TB17F	1.6861915486031354
+1PPF_A	1PPF_B	PB14C	-1.9771083178293
+1PPF_A	1PPF_B	YB20D	6.330163614091897
+1PPF_A	1PPF_B	LB18G	3.286507401252546
+1PPF_A	1PPF_B	NB36L	2.6103170641009843
+1PPF_A	1PPF_B	EB19L	1.056178691469162
+1PPF_A	1PPF_B	KB13Q	0.2462978387934616
+1PPF_A	1PPF_B	RB21S	0.4137788028550453
+1PPF_A	1PPF_B	RB21W	-0.525587222200782
+1PPF_A	1PPF_B	NB36R	1.843730499681236
+1PPF_A	1PPF_B	EB19I	0.7132576401854376
+1PPF_A	1PPF_B	LB18W	7.439172535222286
+1PPF_A	1PPF_B	PB14W	-1.9771083178293
+1PPF_A	1PPF_B	RB21N	0.3248915324240951
+1PPF_A	1PPF_B	LB18T	0.902728745416976
+1PPF_A	1PPF_B	TB17R	3.403623598200179
+1PPF_A	1PPF_B	EB19M	1.132726128909928
+1PPF_A	1PPF_B	GB32F	1.4611191178070104
+1PPF_A	1PPF_B	PB14S	-0.5635830138822371
+1PPF_A	1PPF_B	EB19P	3.165696958334093
+1PPF_A	1PPF_B	NB36T	0.545192178281269
+1PPF_A	1PPF_B	NB36D	-3.0351094045589195
+1PPF_A	1PPF_B	GB32K	3.2602465512109724
+1PPF_A	1PPF_B	GB32C	1.162691955909022
+1PPF_A	1PPF_B	PB14R	-0.1357296294440111
+1PPF_A	1PPF_B	EB19T	1.4840320759073862
+1PPF_A	1PPF_B	KB13I	0.6225010517184497
+1PPF_A	1PPF_B	NB36A	-1.622618461874907
+1PPF_A	1PPF_B	YB20L	1.3557730976014764
+1PPF_A	1PPF_B	AB15D	1.1223859378254897
+1PPF_A	1PPF_B	GB32T	2.773253383796048
+1R0R_A	1R0R_B	NB31V	0.6594309628674253
+1R0R_A	1R0R_B	NB31A	-0.0326913557481756
+1R0R_A	1R0R_B	EB14A	2.060353100880766
+1R0R_A	1R0R_B	EB14C	2.338571748937099
+1R0R_A	1R0R_B	KB8D	-0.5850517901584239
+1R0R_A	1R0R_B	PB9L	1.1677700289677873
+1R0R_A	1R0R_B	GB27D	2.84790490909284
+1R0R_A	1R0R_B	YB15M	3.0953063843525896
+1R0R_A	1R0R_B	TB12L	0.6594309628674253
+1R0R_A	1R0R_B	LB13H	0.5617201758212111
+1R0R_A	1R0R_B	LB13N	1.47950540973096
+1R0R_A	1R0R_B	RB16C	0.1134270202185145
+1R0R_A	1R0R_B	KB8P	1.3290659418356991
+1R0R_A	1R0R_B	RB16H	0.5183674465563648
+1R0R_A	1R0R_B	RB16Y	0.8118645745984789
+1R0R_A	1R0R_B	RB16T	0.3398448940085981
+1R0R_A	1R0R_B	AB10C	-0.4767381990397759
+1R0R_A	1R0R_B	EB14S	3.1309573535441046
+1R0R_A	1R0R_B	EB14N	2.088856585326744
+1R0R_A	1R0R_B	GB27C	1.4001406877808158
+1R0R_A	1R0R_B	KB8Y	-0.6857520161399435
+1R0R_A	1R0R_B	PB9V	1.3124916232591932
+1R0R_A	1R0R_B	PB9E	-2.2117787708039143
+1R0R_A	1R0R_B	EB14K	2.6742489208430658
+1R0R_A	1R0R_B	AB10Q	1.7857743745367376
+1R0R_A	1R0R_B	KB8R	0.1343224307235893
+1R0R_A	1R0R_B	KB8T	0.8773641218435113
+1R0R_A	1R0R_B	RB16I	0.2041836086855024
+1R0R_A	1R0R_B	PB9H	-0.213503626125453
+1R0R_A	1R0R_B	AB10G	1.0962621126572838
+1R0R_A	1R0R_B	TB12F	-0.0650955323292521
+1R0R_A	1R0R_B	LB13F	0.6084231284666526
+1R0R_A	1R0R_B	GB27W	1.3817679300843615
+1R0R_A	1R0R_B	NB31T	0.5617201758212111
+1R0R_A	1R0R_B	EB14W	1.8237701662181929
+1R0R_A	1R0R_B	YB15A	5.398347488125372
+1R0R_A	1R0R_B	TB12G	3.0953063843525896
+1R0R_A	1R0R_B	RB16Q	0.1798679955417252
+1R0R_A	1R0R_B	TB12D	2.84790490909284
+1R0R_A	1R0R_B	RB16A	-0.0947107308023032
+1R0R_A	1R0R_B	EB14M	1.5493665876928733
+1R0R_A	1R0R_B	LB13P	7.580720015497826
+1R0R_A	1R0R_B	PB9S	0.1343224307235893
+1R0R_A	1R0R_B	LB13S	1.5493665876928733
+1R0R_A	1R0R_B	GB27S	0.8990843016924277
+1R0R_A	1R0R_B	KB8W	-1.0900501053653286
+1R0R_A	1R0R_B	GB27Y	0.4785871872108221
+1R0R_A	1R0R_B	YB15L	2.539273024063462
+1R0R_A	1R0R_B	RB16D	-0.1774129500395815
+1R0R_A	1R0R_B	TB12C	0.8018779118424888
+1R0R_A	1R0R_B	PB9Q	-1.3949326428341244
+1R0R_A	1R0R_B	EB14V	0.0934522353437419
+1R0R_A	1R0R_B	NB31R	1.4386352143511107
+1R0R_A	1R0R_B	AB10E	4.51413612423293
+1R0R_A	1R0R_B	GB27R	3.422804496059225
+1R0R_A	1R0R_B	KB8V	0.404940426337852
+1R0R_A	1R0R_B	AB10F	-2.2117787708039143
+1R0R_A	1R0R_B	EB14D	0.255132873642042
+1R0R_A	1R0R_B	AB10M	-1.7154079136072866
+1R0R_A	1R0R_B	LB13C	-1.3275674069114751
+1R0R_A	1R0R_B	YB15K	5.435689624802924
+1R0R_A	1R0R_B	TB12M	1.3124916232591932
+1R0R_A	1R0R_B	NB31P	1.4586099992258852
+1R0R_A	1R0R_B	TB12P	4.8810807588893255
+1R0R_A	1R0R_B	AB10S	1.7174320495970417
+1R0R_A	1R0R_B	AB10N	2.0661661393391064
+1R0R_A	1R0R_B	GB27H	2.161977587461724
+1R0R_A	1R0R_B	KB8I	1.1083080145736908
+1R0R_A	1R0R_B	RB16N	-0.0497496638267538
+1R0R_A	1R0R_B	RB16M	0.440591395529367
+1R0R_A	1R0R_B	PB9N	-0.9045915834780036
+1R0R_A	1R0R_B	KB8A	-0.6015093726879961
+1R0R_A	1R0R_B	LB13V	2.1265603472370938
+1R0R_A	1R0R_B	NB31G	0.2814924821226849
+1R0R_A	1R0R_B	AB10T	1.2504722482050656
+1R0R_A	1R0R_B	NB31F	-0.1774129500395815
+1R0R_A	1R0R_B	KB8G	-0.8268155324510058
+1R0R_A	1R0R_B	AB10R	5.453502149288758
+1R0R_A	1R0R_B	PB9Y	0.8455318218672154
+1R0R_A	1R0R_B	TB12R	3.2987890864813925
+1R0R_A	1R0R_B	YB15P	6.561810163862448
+1R0R_A	1R0R_B	RB16L	0.2288720236004682
+1R0R_A	1R0R_B	NB31H	0.2814924821226849
+1R0R_A	1R0R_B	AB10H	1.750123405345219
+1R0R_A	1R0R_B	LB13Y	0.440591395529367
+1R0R_A	1R0R_B	LB13T	0.1575389510780986
+1R0R_A	1R0R_B	EB14F	3.567730079858249
+1R0R_A	1R0R_B	AB10L	-1.7305829367964058
+1R0R_A	1R0R_B	GB27I	1.9536061074740232
+1R0R_A	1R0R_B	PB9R	1.0870618351929124
+1R0R_A	1R0R_B	AB10K	3.030210852023337
+1R0R_A	1R0R_B	KB8L	-0.3234635409455801
+1R0R_A	1R0R_B	AB10Y	-0.8268155324510058
+1R0R_A	1R0R_B	LB13W	1.5027219300854693
+1R0R_A	1R0R_B	EB14T	4.193087888100207
+1R0R_A	1R0R_B	NB31L	1.525050974549096
+1R0R_A	1R0R_B	GB27L	2.3771819587328427
+1R0R_A	1R0R_B	NB31S	0.4785871872108221
+1R0R_A	1R0R_B	EB14I	0.7267389265440674
+1R0R_A	1R0R_B	GB27E	1.9536061074740232
+1R0R_A	1R0R_B	RB16S	-0.0161170371716714
+1R0R_A	1R0R_B	RB16W	0.9322396118363072
+1R0R_A	1R0R_B	RB16F	0.440591395529367
+1R0R_A	1R0R_B	YB15F	0.6084231284666526
+1R0R_A	1R0R_B	YB15T	5.398347488125372
+1R0R_A	1R0R_B	TB12I	0.255132873642042
+1R0R_A	1R0R_B	EB14H	1.685027873015967
+1R0R_A	1R0R_B	NB31M	0.3102296955355469
+1R0R_A	1R0R_B	KB8C	-0.6070998990335958
+1R0R_A	1R0R_B	GB27F	0.1134270202185145
+1R0R_A	1R0R_B	PB9T	1.1810782478329696
+1R0R_A	1R0R_B	PB9A	-0.6300128571339751
+1R0R_A	1R0R_B	TB12N	0.0176155720958952
+1R0R_A	1R0R_B	TB12K	1.90690315482858
+1R0R_A	1R0R_B	YB15E	4.553916383578473
+1R0R_A	1R0R_B	LB13I	3.208733404571104
+1R0R_A	1R0R_B	AB10I	-1.7744688256547416
+1R0R_A	1R0R_B	YB15N	5.5382708671075775
+1R0R_A	1R0R_B	YB15W	-0.3590968734770712
+1R0R_A	1R0R_B	NB31D	0.5617201758212111
+1R0R_A	1R0R_B	GB27Q	2.004613941874794
+1R0R_A	1R0R_B	YB15R	4.407798007611783
+1R0R_A	1R0R_B	EB14Q	1.264327574561456
+1R0R_A	1R0R_B	GB27P	1.0650137263177406
+1R0R_A	1R0R_B	NB31W	0.0549577087734469
+1R0R_A	1R0R_B	NB31Q	-0.2706179956142591
+1R0R_A	1R0R_B	TB12H	0.5183674465563648
+1R0R_A	1R0R_B	KB8E	-0.1089026565024937
+1R0R_A	1R0R_B	EB14P	3.712085396076529
+1R0R_A	1R0R_B	PB9D	-2.4861823493292365
+1R0R_A	1R0R_B	KB8Q	-0.5234995926941153
+1R0R_A	1R0R_B	EB14Y	1.863550425563736
+1R0R_A	1R0R_B	EB14G	2.6742489208430658
+1R0R_A	1R0R_B	YB15C	3.208733404571104
+1R0R_A	1R0R_B	KB8M	-0.5534654196932056
+1R0R_A	1R0R_B	KB8S	-0.3030822107900057
+1R0R_A	1R0R_B	NB31K	1.1316793528819176
+1R0R_A	1R0R_B	NB31I	0.3102296955355469
+1R0R_A	1R0R_B	NB31E	-0.2021594726957474
+1R0R_A	1R0R_B	LB13G	2.3201410234148216
+1R0R_A	1R0R_B	AB10W	0.2041836086855024
+1R0R_A	1R0R_B	PB9C	-0.972933908417696
+1R0R_A	1R0R_B	KB8N	-0.3853999577890352
+1R0R_A	1R0R_B	LB13R	2.894549566700244
+1R0R_A	1R0R_B	KB8H	-0.5850517901584239
+1R0R_A	1R0R_B	TB12Q	2.1907148008745856
+1R0R_A	1R0R_B	PB9M	-0.0326913557481756
+1R0R_A	1R0R_B	YB15V	3.4612168154183887
+1R0R_A	1R0R_B	EB14R	2.0661661393391064
+1R0R_A	1R0R_B	LB13D	4.476140332551474
+1R0R_A	1R0R_B	PB9K	0.3398448940085981
+1R0R_A	1R0R_B	RB16G	2.060353100880766
+1R0R_A	1R0R_B	YB15S	5.0908526523600806
+1R0R_A	1R0R_B	PB9I	1.7174320495970417
+1R0R_A	1R0R_B	YB15G	6.312971527865785
+1R0R_A	1R0R_B	GB27T	1.655412674542916
+1R0R_A	1R0R_B	PB9F	1.362798551103266
+1R0R_A	1R0R_B	AB10P	3.2987890864813925
+1R0R_A	1R0R_B	GB27N	1.264327574561456
+1R0R_A	1R0R_B	KB8F	-0.7834628031861595
+1R0R_A	1R0R_B	NB31Y	0.1798679955417252
+1R0R_A	1R0R_B	RB16E	0.3398448940085981
+1R0R_A	1R0R_B	TB12W	2.894549566700244
+1R0R_A	1R0R_B	PB9G	-0.3410085695957647
+1R0R_A	1R0R_B	GB27K	2.894549566700244
+1R0R_A	1R0R_B	RB16K	-0.0947107308023032
+1R0R_A	1R0R_B	AB10D	5.155659485233353
+1R0R_A	1R0R_B	YB15Q	4.440489363359959
+1R0R_A	1R0R_B	PB9W	0.4785871872108221
+1R0R_A	1R0R_B	GB27M	2.1265603472370938
+1R0R_A	1R0R_B	NB31C	0.5617201758212111
+1R0R_A	1R0R_B	RB16V	0.1134270202185145
+1R0R_A	1R0R_B	EB14L	0.5617201758212111
+1R0R_A	1R0R_B	RB16P	7.287635070858058
+1R0R_A	1R0R_B	YB15I	2.8246883887383287
+1R0R_A	1R0R_B	TB12Y	-0.8665957917965468
+1S0W_A	1S0W_B	AB142F	-2.1019341179373274
+1S1Q_A	1S1Q_B	FA86A	0.7618593974133088
+1S1Q_A	1S1Q_B	NA43A	1.210689233846836
+1S1Q_A	1S1Q_B	WA73A	0.2759324833310703
+1S1Q_A	1S1Q_B	FA42A	0.1959003444918128
+1S1Q_A	1S1Q_B	DA44A	0.9488435117371492
+1S1Q_A	1S1Q_B	VA41A	0.6591666674866832
+1SBB_A	1SBB_B	VB26Y	-1.4547657175287825
+1SBB_A	1SBB_B	LB20T	-0.0912809292832177
+1SBB_A	1SBB_B	YB91V	0.0790711375448189
+1SBN_A	1SBN_B	RB38K	-0.3098434393028242
+1SGD_A	1SGD_B	DB13L	-5.589755775252961
+1SGE_A	1SGE_B	EB13L	-5.899483203299344
+1SGN_A	1SGN_B	NB13L	-3.3524942922738106
+1SGP_A	1SGP_B	AB13L	-2.9475538659359604
+1SGQ_A	1SGQ_B	GB13L	-4.933851329158159
+1SGY_A	1SGY_B	YB13L	-1.6526778147726642
+1SIB_A	1SIB_B	KB46R	-0.7311587735816741
+1TM1_A	1TM1_B	MB40A	1.0271508140535437
+1TM1_A	1TM1_B	MB40F	1.0271508140535437
+1TM1_A	1TM1_B	TB39D	2.0980176673647684
+1TM1_A	1TM1_B	RB48C	3.234173525639145
+1TM1_A	1TM1_B	MB40K	1.093013509854524
+1TM1_A	1TM1_B	YB42G	4.678612295367822
+1TM1_A	1TM1_B	EB41S	2.74638619860184
+1TM1_A	1TM1_B	MB40Y	0.0564382965578786
+1TM1_A	1TM1_B	TB39P	3.754120390544673
+1TM1_A	1TM1_B	RB43A	1.2555223637452997
+1TM1_A	1TM1_B	MB40G	2.231782031124311
+1TM3_A	1TM3_B	KB40M	-1.093013509854524
+1TM4_A	1TM4_B	GB40M	-2.231782031124311
+1TM5_A	1TM5_B	AB40M	-1.0271508140535437
+1TM7_A	1TM7_B	YB40M	-0.0564382965578786
+1TMG_A	1TMG_B	FB40M	-1.0271508140535437
+1TO1_A	1TO1_B	AB42Y	-2.980082225320676
+1UUZ_A	1UUZ_B	HA62D	3.3775171494500267
+1UUZ_A	1UUZ_B	HA62N	1.5188448158376406
+1UUZ_A	1UUZ_B	HA62A	1.7739346081390313
+1UUZ_A	1UUZ_B	CA64A	0.6505475729588959
+1UUZ_A	1UUZ_B	HA62Q	1.5188448158376406
+1X1X_A	1X1X_B	AB76E	-0.8231384096358614
+1XD3_A	1XD3_B	RB42L	-0.8819965836107677
+1XD3_A	1XD3_B	IB44A	0.2720923755953315
+1XD3_A	1XD3_B	KB11R	1.4346274012921905
+1XD3_A	1XD3_B	LB8A	2.738291482575054
+1XD3_A	1XD3_B	DB39A	-0.4200912586157965
+1XD3_A	1XD3_B	KB6A	1.37496645878797
+1XD3_A	1XD3_B	RB72L	1.33165778357358
+1XD3_A	1XD3_B	KB27R	0.2720923755953315
+1XD3_A	1XD3_B	HB68N	0.0
+1XD3_A	1XD3_B	KB27A	-0.0638551708388259
+1XD3_A	1XD3_B	RB74L	2.4323607052688656
+1XD3_A	1XD3_B	KB6R	0.309592804009819
+1Y1K_A	1Y1K_B	AB39T	-2.726969576491131
+1Y33_A	1Y33_B	PB39T	-3.754120390544673
+1Y34_A	1Y34_B	AB41E	-3.0526816709112428
+1Y3B_A	1Y3B_B	SB41E	-2.74638619860184
+1Z7X_A	1Z7X_B	WA263A	2.211305151040648
+1Z7X_A	1Z7X_B	YA434A	5.952934775265284
+1Z7X_A	1Z7X_B	EA287A	1.320659825016321
+1Z7X_A	1Z7X_B	SA289A	0.8139737839962642
+1Z7X_A	1Z7X_B	IA459A	0.3372513313386136
+1Z7X_A	1Z7X_B	RA457A	0.8478203627950833
+1Z7X_A	1Z7X_B	YA437F	2.1649678059217905
+1Z7X_A	1Z7X_B	YA437A	2.6228152289070703
+1Z7X_A	1Z7X_B	WA261A	1.3352816523491953
+1Z7X_A	1Z7X_B	EA206A	1.0181724296231174
+1Z7X_A	1Z7X_B	WA318A	0.9929705871243611
+1Z7X_A	1Z7X_B	YA434F	0.1205619193917186
+1Z7X_A	1Z7X_B	WA375A	1.6687200025820168
+1Z7X_A	1Z7X_B	DA435A	3.660587922732521
+1Z7X_A	1Z7X_B	EA344A	1.560757578081752
+1Z7X_A	1Z7X_B	KA320A	1.320659825016321
+1Z7X_A	1Z7X_B	EA401A	1.3056677857803614
+2A9K_A	2A9K_B	GB55D	2.409802431359984
+2AJF_A	2AJF_B	TB159S	1.8230168607872856
+2AJF_A	2AJF_B	NB151K	2.0115904859103293
+2B0Z_A	2B0Z_B	IB87F	-2.527726222707648
+2B10_A	2B10_B	SB87F	-1.1642414344620822
+2B11_A	2B11_B	WB87F	-0.7537916145686152
+2B12_A	2B12_B	YB87F	-1.574691254355549
+2B42_A	2B42_B	HA357K	2.5406589081690485
+2B42_A	2B42_B	HA357A	1.6211478042605787
+2B42_A	2B42_B	HA357Q	1.064746296952702
+2BTF_A	2BTF_B	GB120F	1.7200653130929708
+2BTF_A	2BTF_B	KB125A	0.4598245664571241
+2BTF_A	2BTF_B	FB59A	1.5949378050109324
+2BTF_A	2BTF_B	VB60E	0.9782368108508566
+2FTL_A	2FTL_B	KB15D	11.38118981571144
+2FTL_A	2FTL_B	GB36A	2.190994415843768
+2FTL_A	2FTL_B	KB15S	7.64231393280437
+2FTL_A	2FTL_B	KB15G	12.221723141911871
+2FTL_A	2FTL_B	KB15N	7.948233485706734
+2FTL_A	2FTL_B	KB15H	8.689000456471591
+2FTL_A	2FTL_B	KB15E	9.324441184036544
+2FTL_A	2FTL_B	KB15Q	9.271641744362888
+2FTL_A	2FTL_B	IB18A	4.968786397859223
+2FTL_A	2FTL_B	KB15M	7.611102390346371
+2FTL_A	2FTL_B	KB15Y	6.7845643875760775
+2FTL_A	2FTL_B	KB15F	6.952966976747844
+2FTL_A	2FTL_B	KB15I	11.053423601134748
+2FTL_A	2FTL_B	KB15L	8.770452779504572
+2FTL_A	2FTL_B	GB12A	4.346368041583864
+2FTL_A	2FTL_B	KB15T	10.4855197758984
+2FTL_A	2FTL_B	KB15V	11.632235033424028
+2FTL_A	2FTL_B	KB15W	8.572581166725215
+2G2U_A	2G2U_B	DA79K	-0.4526592630612107
+2G2U_A	2G2U_B	RB160A	0.6694568519687198
+2G2U_A	2G2U_B	GB48A	-0.4263831736089845
+2G2U_A	2G2U_B	FB142A	0.2756885272923011
+2G2U_A	2G2U_B	EB31A	0.6505475729588968
+2G2U_A	2G2U_B	WB112A	0.9582522172533592
+2G2U_A	2G2U_B	HB148A	1.1183877055633626
+2G2U_A	2G2U_B	SB39A	-0.9556609323880992
+2G2U_A	2G2U_B	EB73M	-3.5342558829494286
+2G2U_A	2G2U_B	HB41A	1.7161847851985126
+2G2U_A	2G2U_B	YB143A	-1.846306258412426
+2G2U_A	2G2U_B	KB74A	-0.2172258716374013
+2G2U_A	2G2U_B	FB36A	2.762560350718
+2G2U_A	2G2U_B	YB50A	-2.0746778081041817
+2G2U_A	2G2U_B	WB162A	0.5309427839523906
+2G2U_A	2G2U_B	SB35A	-0.9504205980140484
+2G2U_A	2G2U_B	GB141A	-0.4136023770452688
+2G2U_A	2G2U_B	SB113A	-0.6120034064766262
+2G2U_A	2G2U_B	YB51A	-0.6284324378571107
+2G2U_A	2G2U_B	SB71A	-0.5119496660751359
+2G2U_A	2G2U_B	YB53A	2.3005864028821463
+2G2U_A	2G2U_B	RB144A	-0.3422199365820422
+2G2U_A	2G2U_B	WB150A	1.7843235398671444
+2G2U_A	2G2U_B	EB73A	-1.978441586977592
+2G2W_A	2G2W_B	KA79D	0.4526592630612107
+2GYK_A	2GYK_B	AA49D	-5.915151095926328
+2HLE_A	2HLE_B	KA141Q	-0.4104498198934649
+2HLE_A	2HLE_B	LA87R	2.286146231272053
+2HRK_A	2HRK_B	TA108V	0.7207069189544306
+2HRK_A	2HRK_B	KA140A	0.9530349683520996
+2HRK_A	2HRK_B	TB55V	0.6238866235542542
+2I26_A	2I26_B	AA29V	0.0303735253256078
+2I9B_A	2I9B_B	RB129A	-0.287495356157244
+2I9B_A	2I9B_B	KB131A	0.6739390994047731
+2J0T_A	2J0T_B	SB68A	2.105313290487679
+2J0T_A	2J0T_B	VB69I	0.8090955198345746
+2J0T_A	2J0T_B	SB68R	2.834932000991396
+2J0T_A	2J0T_B	SB68Y	2.976855220496777
+2J0T_A	2J0T_B	VB4K	1.8236053576679616
+2J0T_A	2J0T_B	TB2A	4.387569519641223
+2J0T_A	2J0T_B	CB70S	4.407535772449172
+2J0T_A	2J0T_B	VB4I	1.641799279573867
+2J0T_A	2J0T_B	TB2L	2.6808043441244465
+2J0T_A	2J0T_B	SB68E	2.1843844280325
+2J0T_A	2J0T_B	VB4A	0.0
+2J0T_A	2J0T_B	MB66A	1.6803650344631844
+2J0T_A	2J0T_B	VB4S	0.9819262553075436
+2J0T_A	2J0T_B	VB69T	-0.0774751502088015
+2J0T_A	2J0T_B	TB2R	5.043489333088793
+2J0T_A	2J0T_B	TB2S	1.5936301403211883
+2J1K_A	2J1K_B	GA10D	0.1836594565075522
+2J1K_A	2J1K_B	RA24A	0.7644613432290566
+2NU0_A	2NU0_B	WB13L	-1.839245851362272
+2NU1_A	2NU1_B	HB13L	-1.690019951450214
+2NU2_A	2NU2_B	RB13L	-3.3200901156927367
+2NU4_A	2NU4_B	KB13L	-3.137784172762238
+2O3B_A	2O3B_B	DB75N	5.900288080314299
+2O3B_A	2O3B_B	EB24A	5.472160711277267
+2O3B_A	2O3B_B	EB24Q	5.395647478010924
+2O3B_A	2O3B_B	EB24D	0.5990234450165097
+2O3B_A	2O3B_B	DB75E	5.435139036830098
+2O3B_A	2O3B_B	QB74A	3.231337986686892
+2O3B_A	2O3B_B	WB76A	4.071654248584531
+2OOB_A	2OOB_B	GA13S	0.2352455349568441
+2PCB_A	2PCB_B	EA35Q	0.6756671661458079
+2PCB_A	2PCB_B	AA193F	0.7333081549604046
+2PCB_A	2PCB_B	EA290N	0.8705703893158621
+2PCB_A	2PCB_B	DA34N	0.8208996397869335
+2PCB_A	2PCB_B	EA32Q	0.6054483245339028
+2PCC_A	2PCC_B	EA290A	6.19943425467731
+2PCC_A	2PCC_B	DA34A	-0.8965966717942191
+2PCC_A	2PCC_B	VA197A	2.101623846030991
+2PCC_A	2PCC_B	AB86G	1.900898253877345
+2PCC_A	2PCC_B	KB92A	0.9014172118349733
+2PCC_A	2PCC_B	KB77A	0.3035163338270603
+2SGP_A	2SGP_B	PB13L	-8.328285781965437
+2SGQ_A	2SGQ_B	QB13L	-2.567301854513074
+2SIC_A	2SIC_B	MB67K	0.0
+2SIC_A	2SIC_B	MB67L	-0.2400977530654327
+2SIC_A	2SIC_B	MB67V	0.7129372152866669
+2SIC_A	2SIC_B	MB67G	0.1451502243292779
+2SIC_A	2SIC_B	MB67E	0.7956977972881719
+2SIC_A	2SIC_B	MB67I	1.6035825413109936
+2SIC_A	2SIC_B	MB67A	0.2177496699198453
+2SIC_A	2SIC_B	MB67H	0.2177496699198453
+2SIC_A	2SIC_B	MB67D	0.7511539535495384
+2SIC_A	2SIC_B	MB67R	0.0
+2VLN_A	2VLN_B	AB75N	-2.3344109640223465
+2VLO_A	2VLO_B	AB95K	-1.959528419718087
+2VLQ_A	2VLQ_B	AB86F	-3.8787286930522846
+2WPT_A	2WPT_B	VA34A	3.8073835914542
+2WPT_A	2WPT_B	RA35T	-1.0377726938254526
+2WPT_A	2WPT_B	SA47A	2.4244821810979267
+2WPT_A	2WPT_B	PA53A	2.926212930274612
+2WPT_A	2WPT_B	NA31A	-0.3796816568193666
+2WPT_A	2WPT_B	NA31V	-0.8965966717942226
+2WPT_A	2WPT_B	EA27A	1.7330802088570818
+2WPT_A	2WPT_B	DA30L	-3.4016900535149284
+2WPT_A	2WPT_B	EA38A	4.500904184630159
+2WPT_A	2WPT_B	DA30A	-0.132135328565166
+2WPT_A	2WPT_B	RA35A	-1.1103721394160235
+3BK3_A	3BK3_B	LB2R	0.02632228544363
+3BK3_A	3BK3_B	IB22R	2.1204975562858497
+3BK3_A	3BK3_B	TB6P	1.7421427493935546
+3BK3_A	3BK3_B	LB2A	0.0
+3BK3_A	3BK3_B	IB28R	1.1749111631225233
+3BK3_A	3BK3_B	IB28A	1.2608812593210048
+3BK3_A	3BK3_B	IB19R	0.4237572095729831
+3BK3_A	3BK3_B	IB3A	1.0381314662809549
+3BK3_A	3BK3_B	IB3R	0.6505475729588959
+3BK3_A	3BK3_B	AB37R	1.3820325122991317
+3BK3_A	3BK3_B	IB19A	0.4861468519007541
+3BK3_A	3BK3_B	TB4P	0.4861468519007541
+3BK3_A	3BK3_B	IB22A	1.3070464916876876
+3BP8_A	3BP8_B	FA115A	0.7086306223429606
+3BP8_A	3BP8_B	AB51F	0.615184594049202
+3BTD_A	3BTD_B	DB13K	-11.38118981571144
+3BTE_A	3BTE_B	EB13K	-9.324441184036544
+3BTF_A	3BTF_B	FB13K	-6.952966976747844
+3BTG_A	3BTG_B	GB13K	-12.221723141911871
+3BTH_A	3BTH_B	HB13K	-8.689000456471591
+3BTM_A	3BTM_B	MB13K	-7.611102390346371
+3BTQ_A	3BTQ_B	QB13K	-9.271641744362888
+3BTT_A	3BTT_B	TB13K	-10.4855197758984
+3BTW_A	3BTW_B	WB13K	-8.572581166725215
+3SGB_A	3SGB_B	NB30T	0.0097100109797683
+3SGB_A	3SGB_B	YB14F	0.166431319454233
+3SGB_A	3SGB_B	LB12H	1.690019951450214
+3SGB_A	3SGB_B	NB30S	0.4033062838879129
+3SGB_A	3SGB_B	NB30E	0.9493102265368236
+3SGB_A	3SGB_B	GB26T	2.997860793780033
+3SGB_A	3SGB_B	AB9E	0.6621283342590729
+3SGB_A	3SGB_B	TB11Q	2.899389817238223
+3SGB_A	3SGB_B	LB12G	4.933851329158159
+3SGB_A	3SGB_B	GB26Q	2.9151496893548874
+3SGB_A	3SGB_B	YB14R	2.7221240778696814
+3SGB_A	3SGB_B	LB12K	3.137784172762238
+3SGB_A	3SGB_B	AB9Y	-1.62434887285022
+3SGB_A	3SGB_B	NB30I	0.1951685328670951
+3SGB_A	3SGB_B	TB11N	3.209117245284604
+3SGB_A	3SGB_B	YB14K	3.3851856480219884
+3SGB_A	3SGB_B	LB12V	2.997860793780033
+3SGB_A	3SGB_B	EB13S	1.916554724799452
+3SGB_A	3SGB_B	RB15W	0.3074948357652936
+3SGB_A	3SGB_B	YB14Q	3.235378095326178
+3SGB_A	3SGB_B	AB9L	0.166431319454233
+3SGB_A	3SGB_B	LB12R	3.3200901156927367
+3SGB_A	3SGB_B	KB7T	-1.9218995071345155
+3SGB_A	3SGB_B	TB11K	2.00731131326644
+3SGB_A	3SGB_B	LB12I	4.4188804360352485
+3SGB_A	3SGB_B	YB14S	2.00731131326644
+3SGB_A	3SGB_B	RB15L	0.2247837313401461
+3SGB_A	3SGB_B	AB9F	-1.1642063718404645
+3SGB_A	3SGB_B	YB14T	4.804015387902329
+3SGB_A	3SGB_B	EB13P	2.243485371143421
+3SGB_A	3SGB_B	LB12P	8.328285781965437
+3SGB_A	3SGB_B	PB8M	-0.4935835395167612
+3SGB_A	3SGB_B	KB7Q	-0.6147123198086053
+3SGB_A	3SGB_B	YB14P	6.18897862625524
+3SGB_A	3SGB_B	EB13D	0.5450121373114403
+3SGB_A	3SGB_B	NB30K	0.5713717457920833
+3SGB_A	3SGB_B	GB26W	3.8576093435276473
+3SGB_A	3SGB_B	PB8K	0.1124663136298167
+3SGB_A	3SGB_B	KB7H	-0.3958727524705452
+3SGB_A	3SGB_B	KB7C	-0.5767165281271485
+3SGB_A	3SGB_B	AB9I	-0.4468805868713179
+3SGB_A	3SGB_B	NB30M	0.1387863297181226
+3SGB_A	3SGB_B	KB7F	-0.7154588213293724
+3SGB_A	3SGB_B	LB12F	1.3548929899871391
+3SGB_A	3SGB_B	EB13A	1.0050493855427938
+3SGB_A	3SGB_B	RB15T	0.447418214747497
+3SGB_A	3SGB_B	GB26S	1.5853125788500115
+3SGB_A	3SGB_B	EB13M	-0.1868589353703776
+3SGB_A	3SGB_B	EB13L	0.7684664508802204
+3SGB_A	3SGB_B	KB7V	-0.941876695119456
+3SGB_A	3SGB_B	PB8V	-0.3060925742413527
+3SGB_A	3SGB_B	AB9P	0.5450121373114403
+3SGB_A	3SGB_B	AB9Q	0.4242016943929876
+3SGB_A	3SGB_B	EB13R	1.2572990674231974
+3SGB_A	3SGB_B	GB26N	2.243485371143421
+3SGB_A	3SGB_B	KB7I	-1.3259217109522314
+3SGB_A	3SGB_B	YB14G	2.8028322716445544
+3SGB_A	3SGB_B	YB14A	1.916554724799452
+3SGB_A	3SGB_B	AB9D	0.4697472592111236
+3SGB_A	3SGB_B	RB15A	0.0530042992357202
+3SGB_A	3SGB_B	RB15S	0.2737622264977269
+3SGB_A	3SGB_B	NB30D	0.8983023921360509
+3SGB_A	3SGB_B	RB15V	-0.0233047356125002
+3SGB_A	3SGB_B	PB8D	-1.0187187642609796
+3SGB_A	3SGB_B	AB9W	-1.5320419143777482
+3SGB_A	3SGB_B	PB8N	-0.1230641035016635
+3SGB_A	3SGB_B	YB14C	1.6716471937537598
+3SGB_A	3SGB_B	PB8L	-0.1394065829779744
+3SGB_A	3SGB_B	GB26P	1.0815968229835615
+3SGB_A	3SGB_B	EB13V	0.125774532494999
+3SGB_A	3SGB_B	YB14L	1.5542068382308545
+3SGB_A	3SGB_B	KB7R	-0.6830546447482977
+3SGB_A	3SGB_B	KB7L	-1.7920635658786883
+3SGB_A	3SGB_B	TB11P	3.137784172762238
+3SGB_A	3SGB_B	YB14N	3.114567652407727
+3SGB_A	3SGB_B	EB13I	-0.6147123198086053
+3SGB_A	3SGB_B	LB12T	3.1601132172258635
+3SGB_A	3SGB_B	LB12N	3.3524942922738106
+3SGB_A	3SGB_B	GB26K	2.489867363137748
+3SGB_A	3SGB_B	GB26I	4.035548937022108
+3SGB_A	3SGB_B	GB26Y	3.0530154549434165
+3SGB_A	3SGB_B	AB9K	2.456745628990024
+3SGB_A	3SGB_B	PB8I	-0.0233047356125002
+3SGB_A	3SGB_B	EB13W	1.5261595861742734
+3SGB_A	3SGB_B	AB9M	0.0877197909736509
+3SGB_A	3SGB_B	RB15H	0.447418214747497
+3SGB_A	3SGB_B	RB15Y	0.2898792636693983
+3SGB_A	3SGB_B	RB15D	0.6297241576779964
+3SGB_A	3SGB_B	TB11H	3.45883240889496
+3SGB_A	3SGB_B	NB30H	-0.1230641035016635
+3SGB_A	3SGB_B	RB15C	0.5450121373114403
+3SGB_A	3SGB_B	NB30P	0.5713717457920833
+3SGB_A	3SGB_B	TB11I	1.7285144780205088
+3SGB_A	3SGB_B	TB11D	4.887148376512718
+3SGB_A	3SGB_B	YB14I	3.5419653975053467
+3SGB_A	3SGB_B	NB30V	-0.2385091068836153
+3SGB_A	3SGB_B	EB13G	2.075653638206136
+3SGB_A	3SGB_B	PB8S	0.0648068328732751
+3SGB_A	3SGB_B	KB7S	-2.5515796282966186
+3SGB_A	3SGB_B	RB15P	7.577514334527455
+3SGB_A	3SGB_B	TB11C	2.8855344908818346
+3SGB_A	3SGB_B	PB8Y	-0.3958727524705452
+3SGB_A	3SGB_B	TB11R	1.9452919382123144
+3SGB_A	3SGB_B	RB15E	0.6621283342590729
+3SGB_A	3SGB_B	NB30L	0.7304706591987653
+3SGB_A	3SGB_B	RB15G	1.0917571755118871
+3SGB_A	3SGB_B	PB8T	-0.1071663729985825
+3SGB_A	3SGB_B	PB8Q	-0.3116301090185978
+3SGB_A	3SGB_B	KB7N	-0.6503632890001203
+3SGB_A	3SGB_B	GB26C	2.5753047953462893
+3SGB_A	3SGB_B	AB9G	2.4518568511311223
+3SGB_A	3SGB_B	AB9T	0.3264642147463892
+3SGB_A	3SGB_B	GB26M	2.7667415955586847
+3SGB_A	3SGB_B	AB9H	-0.4468805868713179
+3SGB_A	3SGB_B	GB26F	2.9802452216841377
+3SGB_A	3SGB_B	AB9S	0.8983023921360509
+3SGB_A	3SGB_B	RB15F	0.2247837313401461
+3SGB_A	3SGB_B	PB8C	-0.3228651772934849
+3SGB_A	3SGB_B	TB11Y	3.3200901156927367
+3SGB_A	3SGB_B	NB30F	-0.1394065829779744
+3SGB_A	3SGB_B	EB13N	1.086699087248979
+3SGB_A	3SGB_B	LB12D	5.589755775252961
+3SGB_A	3SGB_B	KB7D	-0.6147123198086053
+3SGB_A	3SGB_B	TB11L	2.436940154519254
+3SGB_A	3SGB_B	KB7Y	-0.5767165281271485
+3SGB_A	3SGB_B	KB7G	-0.7154588213293724
+3SGB_A	3SGB_B	PB8E	-0.6147123198086053
+3SGB_A	3SGB_B	PB8F	-0.3401335934645768
+3SGB_A	3SGB_B	RB15N	0.3264642147463892
+3SGB_A	3SGB_B	PB8H	0.2737622264977269
+3SGB_A	3SGB_B	AB9C	0.040958397319315
+3SGB_A	3SGB_B	RB15K	-0.2739263471082491
+3SGB_A	3SGB_B	GB26D	1.6023708869285915
+3SGB_A	3SGB_B	TB11W	2.997860793780033
+3SGB_A	3SGB_B	NB30R	0.5713717457920833
+3SGB_A	3SGB_B	PB8W	-0.4935835395167612
+3SGB_A	3SGB_B	KB7A	-2.536338619277537
+3SGB_A	3SGB_B	GB26R	3.45883240889496
+3SGB_A	3SGB_B	YB14W	0.3833314990131402
+3SGB_A	3SGB_B	KB7W	-0.0132029471206074
+3SGB_A	3SGB_B	GB26L	2.6211483482070648
+3SGB_A	3SGB_B	LB12Y	1.6526778147726642
+3SGB_A	3SGB_B	NB30A	0.3264642147463892
+3SGB_A	3SGB_B	PB8A	-0.1868589353703776
+3SGB_A	3SGB_B	TB11G	5.506506448524476
+3SGB_A	3SGB_B	YB14M	1.9749071366853648
+3SGB_A	3SGB_B	NB30Y	-0.1071663729985825
+3SGB_A	3SGB_B	GB26E	1.9452919382123144
+3SGB_A	3SGB_B	TB11M	2.00731131326644
+3SGB_A	3SGB_B	RB15M	0.125774532494999
+3SGB_A	3SGB_B	EB13Y	0.7684664508802204
+3SGB_A	3SGB_B	LB12S	4.100355769895383
+3SGB_A	3SGB_B	RB15Q	0.040958397319315
+3SGB_A	3SGB_B	TB11F	3.5886683501507903
+3SGB_A	3SGB_B	AB9N	-0.0878839115841731
+3SGB_A	3SGB_B	EB13T	2.1136494298875905
+3SGB_A	3SGB_B	RB15I	0.2401295998426445
+3SGB_A	3SGB_B	PB8G	0.0530042992357202
+3SGB_A	3SGB_B	EB13Q	0.1809766071669045
+3SGB_A	3SGB_B	YB14E	2.00731131326644
+3SGB_A	3SGB_B	EB13F	1.916554724799452
+3SGB_A	3SGB_B	EB13H	0.5187512872698665
+3SGB_A	3SGB_B	KB7E	0.0317581198549401
+3SGB_A	3SGB_B	AB9R	2.00731131326644
+3SGB_A	3SGB_B	GB26H	2.700075968994508
+3SGB_A	3SGB_B	NB30C	0.3448369724428453
+3SGB_A	3SGB_B	LB12W	1.839245851362272
+3SGB_A	3SGB_B	NB30Q	0.2401295998426445
+3SGB_A	3SGB_B	YB14V	4.360143692429439
+3SGB_A	3SGB_B	NB30W	0.5187512872698665
+3SGB_A	3SGB_B	PB8R	0.3074948357652936
+3SGB_A	3SGB_B	EB13C	1.1583240436369913
+3SGB_A	3SGB_B	KB7P	0.0
+3SGB_A	3SGB_B	EB13K	0.166431319454233
+3SGB_A	3SGB_B	KB7M	-1.1500144461402757
+3SGB_A	3SGB_B	NB30G	0.3074948357652936
+3SGB_A	3SGB_B	LB12C	-0.0132029471206074
+4CPA_A	4CPA_B	YB36G	0.5017307491766836
+4CPA_A	4CPA_B	VB37I	0.1992433537834816
+4CPA_A	4CPA_B	PB35G	2.5948342479259647
+4CPA_A	4CPA_B	VB37A	2.324362969534924
+4CPA_A	4CPA_B	YB36F	0.0
+4CPA_A	4CPA_B	VB37F	0.0
+4CPA_A	4CPA_B	VB37L	1.2787470627722115
+4CPA_A	4CPA_B	VB37G	3.7328431104389743

examples/sample_mut.txt

+1A22 FPTIPLSRLFDNAMLRAHRLHQLAFDTYQEFEEAYIPKEQKYSFLQNPQTSLCFSESIPTPSNREETQQKSNLELLRISLLLIQSWLEPVQFLRSVFANSLVYGASDSNVYDLLKDLEERIQTLMGRLEGQIFKQTYSKFDTDALLKNYGLLYCFRKDMDKVETFLRIVQCRSVEGSCGF PKFTKCRSPERETFSCHWTLGPIQLFYTRRNTQEWTQEWKECPDYVSAGENSCYFNSSFTSIWIPYCIKLTSNGGTVDEKCFSVDEIVQPDPPIALNWTLLNGIHADIQVRWEAPRNADIQKGWMVLEYELQYKEVNETKWKMMDPILTTSVPVYSLKVDKEYEVRVRSKQRNSGNYGEFSEVLYVTLPQMS CA171A,CB67A,EB12A,FA25A,PA61A,RB11M
\ No newline at end of file

mulan/__init__.py

-from .config import MulanConfig   # noqa
-from .modules import LightAttModel  # noqa
+from .config import MulanConfig   
+from .modules import LightAttModel 
 from .utils import (
     load_pretrained,
     load_pretrained_plm, 
     get_available_models, 
     get_available_plms,
-) # noqa
\ No newline at end of file
+)
\ No newline at end of file

mulan/data.py

+from typing import Dict, List, NamedTuple, Tuple, Optional
+
+import os
+from tqdm import tqdm
+import pandas as pd
+import torch
+from torch.utils.data import Dataset, default_collate
+from torch.nn.utils.rnn import pad_sequence
+import numpy as np
+
+from mulan import utils
+
+
+class MutatedComplex(NamedTuple):
+    sequence_A: str
+    sequence_B: str
+    mutations: Tuple[str]
+
+
+class MutatedComplexEmbeds(NamedTuple):
+    seq1: torch.Tensor
+    seq2: torch.Tensor
+    mut_seq1: torch.Tensor
+    mut_seq2: torch.Tensor
+
+
+class MulanDataset(Dataset):
+    def __init__(
+        self,
+        mutated_complexes: List[MutatedComplex],
+        wt_sequences: Dict[str, str],
+        embeddings_dir: str,
+        plm_model_name: str = None,
+        scores: List[float] = None,
+        zs_scores: List[float] = None,
+    ):
+
+        self.sequences = wt_sequences
+        self.embeddings_dir = embeddings_dir
+        self.mutated_complexes = mutated_complexes
+        self.zs_scores = zs_scores
+        self.scores = scores
+        self._sequences_ids = []
+        self._fill_metadata(mutated_complexes)
+
+        # generate embeddings if not provided
+        all_ids = set([id_ for ids in self._sequences_ids for id_ in ids])
+        provided_embeddings_ids = (
+            [os.path.splitext(file)[0] for file in os.listdir(self.embeddings_dir)]
+            if os.path.exists(self.embeddings_dir)
+            else []
+        )
+        missing_ids = all_ids - set(provided_embeddings_ids)
+        if missing_ids:
+            if not plm_model_name:
+                raise ValueError(
+                    "`plm_model_name` must be provided if embeddings were not pre-computed."
+                )
+            self._generate_missing_embeddings(plm_model_name, missing_ids)
+
+    def __len__(self):
+        return len(self.mutated_complexes)
+
+    def __getitem__(self, index):
+        return {
+            "data": self.mutated_complexes[index],
+            "inputs_embeds": self._load_embeddings(index),
+            "zs_scores": (
+                torch.tensor(self.zs_scores[index], dtype=torch.float32)
+                if self.zs_scores
+                else None
+            ),
+            "labels": (
+                torch.tensor(self.scores[index], dtype=torch.float32) if self.scores else None
+            ),
+        }
+
+    @classmethod
+    def from_table(
+        cls,
+        mutated_complexes_file: str,
+        wt_sequences_file: str,
+        embeddings_dir: str,
+        plm_model_name: str = None,
+    ):
+        wt_sequences = utils.parse_fasta(wt_sequences_file)
+        # parse table file
+        data = pd.read_table(mutated_complexes_file, sep=r"\s+", header=None)
+        mutated_complexes = [
+            MutatedComplex(row[0], row[1], tuple(row[2].split(",")))
+            for row in data.itertuples(index=False)
+        ]
+        scores, zs_scores = None, None
+        if len(data.columns) > 3:
+            scores = data[3].astype(float).tolist()
+        if len(data.columns) > 4:
+            zs_scores = data[4].astype(float).tolist()
+        return cls(
+            mutated_complexes, wt_sequences, embeddings_dir, plm_model_name, scores, zs_scores
+        )
+
+    def _fill_metadata(self, mutated_complexes):
+        for seq1_label, seq2_label, mutations in mutated_complexes:
+            seq1 = self.sequences[seq1_label]
+            seq2 = self.sequences[seq2_label]
+            mut_seq1, mut_seq2 = utils.parse_mutations(mutations, seq1, seq2)
+            mut_seq1_label = (
+                f"{seq1_label}_{'-'.join([mut for mut in mutations if mut[1] == 'A'])}"
+            )
+            mut_seq2_label = (
+                f"{seq2_label}_{'-'.join([mut for mut in mutations if mut[1] == 'B'])}"
+            )
+            self.sequences.update({mut_seq1_label: mut_seq1, mut_seq2_label: mut_seq2})
+            self._sequences_ids.append((seq1_label, seq2_label, mut_seq1_label, mut_seq2_label))
+        return
+
+    def _generate_missing_embeddings(self, plm_model_name, missing_ids):
+        plm_model, plm_tokenizer = utils.load_pretrained_plm(plm_model_name)
+        os.makedirs(self.embeddings_dir, exist_ok=True)
+        for id_ in tqdm(missing_ids, desc="Generating embeddings"):
+            seq = self.sequences[id_]
+            embedding = utils.embed_sequence(plm_model, plm_tokenizer, seq)
+            utils.save_embedding(embedding, self.embeddings_dir, id_)
+        # del plm_model, plm_tokenizer
+        return
+
+    def _load_embeddings(self, index):
+        return MutatedComplexEmbeds(
+            *[
+                torch.load(os.path.join(self.embeddings_dir, f"{id_}.pt"), weights_only=True)
+                for id_ in self._sequences_ids[index]
+            ]
+        )
+
+
+class MulanDataCollator(object):
+    def __init__(self, padding_value: float = 0.0):
+        self.padding_value = padding_value
+
+    def __call__(self, batch):
+        return self._collate_fn(batch)
+
+    def _collate_fn(self, batch):
+        elem = batch[0]
+        if isinstance(elem, dict):
+            return {key: self._collate_fn([d[key] for d in batch]) for key in elem}
+        if isinstance(elem, MutatedComplexEmbeds):
+            return MutatedComplexEmbeds(
+                *[
+                    pad_sequence(embeds, batch_first=True, padding_value=self.padding_value)
+                    for embeds in (zip(*batch))
+                ]
+            )
+        elif elem is None:
+            return None
+        else:
+            return default_collate(batch)
+
+
+def split_data(
+    mutated_complexes_file: str,
+    output_dir: Optional[str] = None,
+    add_validation_set: bool = True,
+    validation_size: float = 0.15,
+    test_size: float = 0.15,
+    num_folds: int = 1,
+    random_state: int = 42,
+):
+    """Split data into train, validation and test sets for training or cross-validation."""
+
+    def _save_data(data, output_file):
+        data.to_csv(output_file, sep="\t", index=False, header=False)
+
+    train_data_all, test_data_all = [], []
+    val_data_all = [] if add_validation_set else None
+    files_basename = os.path.splitext(os.path.basename(mutated_complexes_file))[0]
+    data = pd.read_table(mutated_complexes_file, sep=r"\s+", header=None)
+    rng = np.random.default_rng(random_state)
+    if num_folds <= 0:
+        raise ValueError("`num_folds` must be greater than 0.")
+    elif num_folds == 2 and add_validation_set:
+        raise ValueError("`num_folds` must be greater than 2 to add a validation set.")
+    elif num_folds == 1:
+        split_index = rng.choice(
+            [0, 1, 2],
+            size=len(data),
+            p=[test_size, validation_size, 1 - test_size - validation_size],
+        )
+        test_data = data[split_index == 0]
+        if add_validation_set:
+            val_data = data[split_index == 1]
+            train_data = data[split_index == 2]
+            val_data_all.append(val_data)
+        else:
+            train_data = data[(split_index == 1) & (data[split_index == 2])]
+        train_data_all.append(train_data)
+        test_data_all.append(test_data)
+        if output_dir:
+            os.makedirs(output_dir, exist_ok=True)
+            _save_data(train_data, os.path.join(output_dir, f"{files_basename}_train.tsv"))
+            _save_data(test_data, os.path.join(output_dir, f"{files_basename}_test.tsv"))
+            if add_validation_set:
+                _save_data(val_data, os.path.join(output_dir, f"{files_basename}_val.tsv"))
+    else:
+        fold_index = rng.integers(low=0, high=num_folds, size=len(data))
+        for test_fold_index in range(num_folds):
+            test_data = data[fold_index == test_fold_index]
+            if add_validation_set:
+                val_fold_index = (test_fold_index - 1) % num_folds
+                val_data = data[fold_index == val_fold_index]
+                val_data_all.append(val_data)
+                train_data = data[(fold_index != test_fold_index) & (fold_index != val_fold_index)]
+            else:
+                train_data = data[fold_index != test_fold_index]
+            train_data_all.append(train_data)
+            test_data_all.append(test_data)
+            if output_dir:
+                os.makedirs(os.path.join(output_dir, f"fold_{test_fold_index}"), exist_ok=True)
+                _save_data(
+                    train_data,
+                    os.path.join(
+                        output_dir, f"fold_{test_fold_index}", f"{files_basename}_train.tsv"
+                    ),
+                )
+                _save_data(
+                    test_data,
+                    os.path.join(
+                        output_dir, f"fold_{test_fold_index}", f"{files_basename}_test.tsv"
+                    ),
+                )
+                if add_validation_set:
+                    _save_data(
+                        val_data,
+                        os.path.join(
+                            output_dir, f"fold_{test_fold_index}", f"{files_basename}_val.tsv"
+                        ),
+                    )
+    return train_data_all, test_data_all, val_data_all

mulan/train_utils.py

+from typing import Any, Dict, List, Mapping, Optional, Tuple, Union
+from dataclasses import dataclass, field
+import torch
+import torch.nn as nn
+from transformers import Trainer
+from sklearn.metrics import mean_absolute_error, root_mean_squared_error
+from scipy.stats import pearsonr, spearmanr
+
+from mulan.data import MutatedComplexEmbeds, MutatedComplex
+
+
+def _metric_spearmanr(y_true, y_pred):
+    return spearmanr(y_true, y_pred, nan_policy="omit")[0]
+
+
+def _metric_pearsonr(y_true, y_pred):
+    return pearsonr(y_true, y_pred)[0]
+
+
+_DEFAULT_METRICS = {
+    "mae": mean_absolute_error,
+    "rmse": root_mean_squared_error,
+    "pcc": _metric_pearsonr,
+    "scc": _metric_spearmanr,
+}
+
+
+def default_compute_metrics(eval_pred):
+    predictions, labels = eval_pred
+    predictions = predictions.flatten()
+    labels = labels.flatten()
+    res = {}
+    for name, metric in _DEFAULT_METRICS.items():
+        res[name] = metric(labels, predictions)
+    return res
+
+
+@dataclass
+class DatasetArguments:
+    train_data: str = field(
+        metadata={
+            "help": (
+                "Training data in TSV format. Must contain columns for sequence_A, sequence_B,"
+                " mutations (separated with comma if multiple), score and (optionally) zero-shot"
+                " score."
+            )
+        }
+    )
+    train_fasta_file: str = field(
+        metadata={
+            "help": (
+                "Fasta file containing wild-type sequences for training data. Identifiers must"
+                " match the training data and, if present, the evaluation data."
+            )
+        }
+    )
+    embeddings_dir: str = field(
+        metadata={
+            "help": (
+                "Directory containing pre-computed embeddings in PT format, or where new"
+                " embeddings will be stored. In the latter case, `plm_model_name` must be"
+                " provided."
+            )
+        }
+    )
+    eval_data: Optional[str] = field(
+        default=None,
+        metadata={"help": "Evaluation data file, with the same format of training data."},
+    )
+    test_data: Optional[str] = field(
+        default=None,
+        metadata={"help": "Test data file, with the same format of training data."},
+    )
+    test_fasta_file: Optional[str] = field(
+        default=None,
+        metadata={
+            "help": (
+                "Fasta file containing wild-type sequences. Identifiers must match the test data."
+            )
+        },
+    )
+    plm_model_name: Optional[str] = field(
+        default=None,
+        metadata={
+            "help": (
+                "Name of the pre-trained protein language model to use for embedding generation."
+            )
+        },
+    )
+
+
+@dataclass
+class ModelArguments:
+    model_name_or_config_path: str = field(
+        metadata={
+            "help": (
+                "Name of the pre-trained model to fine-tune, or path to config file in JSON"
+                " format."
+            )
+        }
+    )
+    save_model: bool = field(
+        default=False,
+        metadata={"help": "Whether to save the model after training."},
+    )
+
+
+@dataclass
+class CustomisableTrainingArguments:
+    output_dir: str = field(metadata={"help": "Directory where the trained model will be saved."})
+    num_epochs: int = field(default=30, metadata={"help": "Number of training epochs."})
+    batch_size: int = field(default=8, metadata={"help": "Batch size."})
+    learning_rate: float = field(default=5e-4, metadata={"help": "Learning rate."})
+    disable_tqdm: bool = field(
+        default=False, metadata={"help": "Whether to disable tqdm progress bars."}
+    )
+    report_to: Union[None, str, List[str]] = field(
+        default="none",
+        metadata={"help": "The list of integrations to report the results and logs to."},
+    )
+    early_stopping_patience: Optional[int] = field(
+        default=None,
+        metadata={
+            "help": (
+                "Number of epochs without improvement before early stopping. If not set, early"
+                " stopping is disabled."
+            )
+        },
+    )
+
+
+class MulanTrainer(Trainer):
+    """Custom Trainer class adapted for Mulan model training"""
+
+    def compute_loss(self, model, inputs, return_outputs=False):
+        """
+        Computes the loss for MulanDataset inputs for a model that do not return loss values.
+        """
+        inputs.pop("data")
+        outputs = model(inputs["inputs_embeds"], inputs.get("zs_scores"))
+        labels = inputs.get("labels")
+        loss = torch.nn.functional.mse_loss(outputs.view(-1), labels.view(-1))
+        return (loss, outputs) if return_outputs else loss
+
+    def _prepare_input(self, data: Union[torch.Tensor, Any]) -> Union[torch.Tensor, Any]:
+        """
+        Prepares one `data` before feeding it to the model, be it a tensor or a nested list/dictionary of tensors.
+        Adapted from the parent class to handle the case where the input is a custom type.
+        """
+        if isinstance(data, Mapping):
+            return type(data)({k: self._prepare_input(v) for k, v in data.items()})
+        elif isinstance(data, (MutatedComplexEmbeds, MutatedComplex)):
+            return type(data)(*[self._prepare_input(v) for v in data])
+        elif isinstance(data, (tuple, list)):
+            return type(data)(self._prepare_input(v) for v in data)
+        elif isinstance(data, torch.Tensor):
+            kwargs = {"device": self.args.device}
+            if self.is_deepspeed_enabled and (
+                torch.is_floating_point(data) or torch.is_complex(data)
+            ):
+                # NLP models inputs are int/uint and those get adjusted to the right dtype of the
+                # embedding. Other models such as wav2vec2's inputs are already float and thus
+                # may need special handling to match the dtypes of the model
+                kwargs.update(
+                    {"dtype": self.accelerator.state.deepspeed_plugin.hf_ds_config.dtype()}
+                )
+            return data.to(**kwargs)
+        return data
+
+    def prediction_step(
+        self,
+        model: nn.Module,
+        inputs: Dict[str, Union[torch.Tensor, Any]],
+        prediction_loss_only: bool,
+        ignore_keys: Optional[List[str]] = None,
+    ) -> Tuple[Optional[torch.Tensor], Optional[torch.Tensor], Optional[torch.Tensor]]:
+        """
+        Perform an evaluation step on `model` using `inputs`.
+
+        Overridden from the parent class to handle the case where the model does not return loss values.
+        Support for sagemaker was removed!
+
+        Args:
+            model (`nn.Module`):
+                The model to evaluate.
+            inputs (`Dict[str, Union[torch.Tensor, Any]]`):
+                The inputs and targets of the model.
+
+                The dictionary will be unpacked before being fed to the model. Most models expect the targets under the
+                argument `labels`. Check your model's documentation for all accepted arguments.
+            prediction_loss_only (`bool`):
+                Whether or not to return the loss only.
+            ignore_keys (`List[str]`, *optional*):
+                A list of keys in the output of your model (if it is a dictionary) that should be ignored when
+                gathering predictions.
+
+        Return:
+            Tuple[Optional[torch.Tensor], Optional[torch.Tensor], Optional[torch.Tensor]]: A tuple with the loss,
+            logits and labels (each being optional).
+        """
+        has_labels = (
+            False
+            if len(self.label_names) == 0
+            else all(inputs.get(k) is not None for k in self.label_names)
+        )
+        # For CLIP-like models capable of returning loss values.
+        # If `return_loss` is not specified or being `None` in `inputs`, we check if the default value of `return_loss`
+        # is `True` in `model.forward`.
+        return_loss = inputs.get("return_loss", None)
+        if return_loss is None:
+            return_loss = self.can_return_loss
+        # print("return_loss", return_loss, "has_labels", has_labels)
+        loss_without_labels = True if len(self.label_names) == 0 and return_loss else False
+
+        inputs = self._prepare_inputs(inputs)
+        if ignore_keys is None:
+            if hasattr(self.model, "config"):
+                ignore_keys = getattr(self.model.config, "keys_to_ignore_at_inference", [])
+            else:
+                ignore_keys = []
+
+        # labels may be popped when computing the loss (label smoothing for instance) so we grab them first.
+        if has_labels or loss_without_labels:
+            labels = inputs.get("labels")
+        else:
+            labels = None
+
+        with torch.no_grad():
+            if has_labels or loss_without_labels:
+                with self.compute_loss_context_manager():
+                    loss, outputs = self.compute_loss(model, inputs, return_outputs=True)
+                loss = loss.mean().detach()
+                logits = outputs
+            else:
+                loss = None
+                with self.compute_loss_context_manager():
+                    outputs = model(inputs["inputs_embeds"], inputs.get("zs_scores"))
+                logits = outputs
+                # TODO: this needs to be fixed and made cleaner later.
+                if self.args.past_index >= 0:
+                    self._past = outputs[self.args.past_index - 1]
+
+        if prediction_loss_only:
+            return (loss, None, None)
+
+        return (loss, logits, labels)

mulan/utils.py

@@ -9,7 +9,7 @@ import numpy as np
 from scipy.stats import rankdata
 
 import mulan.constants as C
-from mulan.constants import AAs, aa2idx, idx2aa, one2three, three2one  # noqa
+from mulan.constants import AAs, aa2idx, idx2aa, one2three, three2one
 
 
 def mutation_generator(sequence):
@@ -83,7 +83,9 @@ def get_available_models():
     return list(C.MODELS.keys())
 
 
-def load_pretrained_plm(model_name, device="cpu"):
+def load_pretrained_plm(model_name, device=None):
+    if device is None:
+        device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
     model_id = C.PLM_ENCODERS.get(model_name)
     if model_id is None:
         raise ValueError(

requirements.txt

@@ -3,6 +3,8 @@ scipy
 pandas
 tqdm
 h5py
+scikit-learn
 torch>=2.0
 transformers<4.45,>=4.27
+accelerate
 sentencepiece
\ No newline at end of file

scripts/train.py

-"""Train Mulan model on custom data. Not implemented yet!"""
+"""Train Mulan model on custom data using HuggingFace Trainer API"""
+
+import os
+import pandas as pd
+import torch
+from transformers import (
+    HfArgumentParser,
+    TrainingArguments,
+    logging,
+    EarlyStoppingCallback,
+    set_seed,
+)
+
+import mulan
+from mulan.data import MulanDataset, MulanDataCollator
+from mulan.train_utils import (
+    DatasetArguments,
+    ModelArguments,
+    CustomisableTrainingArguments,
+    MulanTrainer,
+    default_compute_metrics,
+)
+
+
+logging.set_verbosity_info()
+logger = logging.get_logger(__name__)
 
-from argparse import ArgumentParser
 
 def get_args():
-    parser = ArgumentParser(
+    parser = HfArgumentParser(
+        dataclass_types=[DatasetArguments, ModelArguments, CustomisableTrainingArguments],
         prog="mulan-train",
         description=__doc__,
     )
-    args = parser.parse_args()
-    return args
+    data_args, model_args, custom_training_args = parser.parse_args_into_dataclasses()
+    return (data_args, model_args, custom_training_args)
+
+
+def load_data(data_args):
+    logger.info("Loading training data")  
+    train_dataset = MulanDataset.from_table(
+        data_args.train_data,
+        data_args.train_fasta_file,
+        data_args.embeddings_dir,
+        data_args.plm_model_name,
+    )
+    eval_dataset, test_dataset = None, None
+    if data_args.eval_data:
+        eval_dataset = MulanDataset.from_table(
+            data_args.eval_data,
+            data_args.train_fasta_file,
+            data_args.embeddings_dir,
+            data_args.plm_model_name,
+        )
+    if data_args.test_data:
+        logger.info("Loading test data...")
+        test_dataset = MulanDataset.from_table(
+            data_args.test_data,
+            data_args.test_fasta_file,
+            data_args.embeddings_dir,
+            data_args.plm_model_name,
+        )
+    return train_dataset, eval_dataset, test_dataset
+
+
+def dummy_forward_call(model, dataset, data_collator):
+    inputs = data_collator([dataset[0]])
+    return model(inputs["inputs_embeds"], inputs.get("zs_scores"))
+
+
+def save_predictions(output_dir, dataset, predictions):
+    df = pd.DataFrame(dataset.mutated_complexes)
+    df["mutations"] = df["mutations"].apply(lambda x: ",".join(x))
+    df["score"] = predictions
+    df.to_csv(
+        os.path.join(output_dir, "test_predictions.tsv"), sep="\t", index=False, header=False
+    )
+    return
+
+
+def save_model_ckpt(model, output_dir):
+    torch.save(
+        {"state_dict": model.state_dict(), "config": model.config.__dict__},
+        os.path.join(output_dir, "model.ckpt"),
+    )
+
+
+def train(data_args, model_args, custom_training_args):
+    # set global seed
+    set_seed(42)
+
+    # load data
+    train_dataset, eval_dataset, test_dataset = load_data(data_args)
+
+    # load model
+    if model_args.model_name_or_config_path in mulan.get_available_models():
+        model = mulan.load_pretrained(model_args.model_name_or_config_path, device="cpu")
+    else:
+        config = mulan.MulanConfig.from_json(model_args.model_name_or_config_path)
+        model = mulan.LightAttModel(config)
+
+    # training arguments
+    training_args = TrainingArguments(
+        output_dir=custom_training_args.output_dir,
+        num_train_epochs=custom_training_args.num_epochs,
+        per_device_train_batch_size=custom_training_args.batch_size,
+        per_device_eval_batch_size=custom_training_args.batch_size,
+        logging_dir=custom_training_args.output_dir,
+        report_to=custom_training_args.report_to,
+        remove_unused_columns=False,
+        label_names=["labels"],
+        logging_strategy="epoch",
+        eval_strategy="epoch" if eval_dataset else "no",
+        save_strategy="epoch" if model_args.save_model else "no",
+        load_best_model_at_end=(eval_dataset and model_args.save_model),  # to be tested
+        metric_for_best_model="loss",
+        save_total_limit=2,
+    )
+
+    data_collator = MulanDataCollator(padding_value=model.config.padding_value)
+    optimizer = torch.optim.AdamW(model.parameters(), lr=custom_training_args.learning_rate)
+    scheduler = torch.optim.lr_scheduler.ReduceLROnPlateau(
+        optimizer, mode="min", factor=0.5, patience=5
+    )
+    early_stopping = (
+        EarlyStoppingCallback(custom_training_args.early_stopping_patience)
+        if custom_training_args.early_stopping_patience
+        else None
+    )
+    dummy_forward_call(model, train_dataset, data_collator)  # to initialize lazy modules
+
+    # instantiate Trainer
+    trainer = MulanTrainer(
+        model=model,
+        args=training_args,
+        data_collator=data_collator,
+        train_dataset=train_dataset,
+        eval_dataset=eval_dataset,
+        compute_metrics=default_compute_metrics,
+        optimizers=(optimizer, scheduler),
+        callbacks=[early_stopping] if early_stopping else None,
+    )
+
+    # train model
+    train_results = trainer.train()
+    metrics = train_results.metrics
+
+    if test_dataset:
+        prediction_results = trainer.predict(test_dataset)
+        save_predictions(
+            custom_training_args.output_dir, test_dataset, prediction_results.predictions
+        )
+        metrics.update(prediction_results.metrics)
+
+    trainer.save_metrics("all", metrics)
+
+    # TODO
+    # remove logging message `Trainer.model is not a `PreTrainedModel`, only saving its state dict.``
+    if model_args.save_model:
+        save_model_ckpt(model, custom_training_args.output_dir)
+
+    return
+
 
 def main():
-    get_args()
-    raise NotImplementedError("This script has not been implemented yet.")
+    data_args, model_args, custom_training_args = get_args()
+    train(data_args, model_args, custom_training_args)
+
 
 if __name__ == "__main__":
     main()
-