Protein-AR
Protein Antibody Reaction
Usage
⁕ Build up required environment first.
python train.py -m attn # you can choose one in {attn,maxfil,lstm,resnet,resnext}
python evaluate -d ./models/
python predict.py -d ./data/sample.txt -m ./models/attns_3.pt
Dataset
x: seq, y: label (0,1) e.g.,)
data preview
| Seq | Label | |
|---|---|---|
| 0 | WSHPSFYPFR | 1 |
| 1 | WLMACFFVFR | 0 |
| 2 | WTVDGLYEYD | 1 |
| 3 | WRATSFYLNT | 0 |
| 4 | WRSIAFFMFA | 0 |
| 5 | YGLRGFYVLT | 1 |
| 6 | WFEFDPYKFR | 0 |
| 7 | WYVFHSFPIL | 0 |
| 8 | WDLYDSYMYT | 0 |
| 9 | FLRISFYVLP | 0 |
| 10 | YFNFHHYLYR | 0 |
data total length
36391
class balance degree
| class | num |
|---|---|
| 0 | 25114 |
| 1 | 11277 |
split info
| train | validation | test |
|---|---|---|
| 21834 | 7278 | 7279 |
path: ./data/split/{train|val|test}.csv
Model
types
model|alias
- ResNet18(full-release)|resnet
- ResNext((full-release)) |resnext
- MaxFilterCNN|maxfil
- LSTM|lstm (w/ OneHot encoding)
- LSTM|lstm0 (w/o OneHot encoding)
- LSTM|lstm1 (w/ emb)
- LSTM|lstm2 (w/ emb+positional emb)
- Self-attention|attns (w/ OneHot encoding)
- Self-attention|attns0 (w/o OneHot encoding)
- Self-attention|attns1 (w/ emb)
- Self-attention|attns2 (w/ emb + positional emb)
- Vision-transformer| vit0 (Original Vit)
- Vision-transformer| vit1 (customized Vit for the small size dataset)
- AutoEncoder | ae1 (w/ classification branch)
Experiment results
-v5 (added ae1)
| Unnamed: 0 | acc | recall | precision | f1 | confusion | loss |
|---|---|---|---|---|---|---|
| maxfiltercnn | 0.836104 | 0.689273 | 0.759648 | 0.722752 | [[4531 492] | 0.00586439 |
| [ 701 1555]] | ||||||
| lstm1 | 0.844209 | 0.738032 | 0.754076 | 0.745968 | [[4480 543] | 0.00568157 |
| [ 591 1665]] | ||||||
| lstm | 0.849292 | 0.744238 | 0.763529 | 0.75376 | [[4503 520] | 0.00563471 |
| [ 577 1679]] | ||||||
| lstm0 | 0.818107 | 0.62234 | 0.748401 | 0.679574 | [[4551 472] | 0.00643181 |
| [ 852 1404]] | ||||||
| vit0 | 0.734854 | 0.659574 | 0.561509 | 0.606604 | [[3861 1162] | 0.0334468 |
| [ 768 1488]] | ||||||
| attns | 0.85259 | 0.74734 | 0.770215 | 0.758605 | [[4520 503] | 0.00551119 |
| [ 570 1686]] | ||||||
| lstm2 | 0.84627 | 0.740691 | 0.757823 | 0.749159 | [[4489 534] | 0.00558329 |
| [ 585 1671]] | ||||||
| resnext | 0.826762 | 0.592642 | 0.796307 | 0.679543 | [[4681 342] | 0.00625525 |
| [ 919 1337]] | ||||||
| ae1 | 0.836104 | 0.714096 | 0.746179 | 0.729785 | [[4475 548] | 0.00585874 |
| [ 645 1611]] | ||||||
| resnet | 0.837752 | 0.651596 | 0.788204 | 0.713419 | [[4628 395] | 0.00599108 |
| [ 786 1470]] | ||||||
| vit1 | 0.819893 | 0.697252 | 0.714675 | 0.705856 | [[4395 628] | 0.0251862 |
| [ 683 1573]] | ||||||
| attns0 | 0.786509 | 0.522606 | 0.711957 | 0.602761 | [[4546 477] | 0.00712849 |
| [1077 1179]] | ||||||
| attns1 | 0.851491 | 0.755319 | 0.763099 | 0.759189 | [[4494 529] | 0.00552442 |
| [ 552 1704]] | ||||||
| attns2 | 0.847644 | 0.75266 | 0.755002 | 0.753829 | [[4472 551] | 0.00557771 |
| [ 558 1698]] |