app.py

import os
import torch
import torch.nn as nn
import pandas as pd
import torch.nn.functional as F
from lavis.models.protein_models.protein_function_opt import Blip2ProteinMistral
from lavis.models.base_model import FAPMConfig
import spaces
import gradio as gr
# from esm_scripts.extract import run_demo
from esm import pretrained, FastaBatchedDataset
from data.evaluate_data.utils import Ontology
import difflib
import re
from transformers import MistralForCausalLM

# Load the trained model
def get_model(type='Molecule Function'):
    model = Blip2ProteinMistral(config=FAPMConfig(), esm_size='3b')
    if type == 'Molecule Function':
        model.load_checkpoint("model/checkpoint_mf2.pth")
        model.Qformer.bert = torch.load('model/mf2_bert.pth', map_location=torch.device('cpu'))
        model.to('cuda')
    elif type == 'Biological Process':
        model.load_checkpoint("model/checkpoint_bp1.pth")
        model.Qformer.bert = torch.load('model/bp1_bert.pth', map_location=torch.device('cpu'))
        model.to('cuda')
    elif type == 'Cellar Component':
        model.load_checkpoint("model/checkpoint_cc2.pth")
        model.Qformer.bert = torch.load('model/cc2_bert.pth', map_location=torch.device('cpu'))
        model.to('cuda')
    return model


models = {
    'Molecule Function': get_model('Molecule Function'),
    'Biological Process': get_model('Biological Process'),
    'Cellular Component': get_model('Cellar Component'),
    }

# Load the mistral model
mistral_model = MistralForCausalLM.from_pretrained("teknium/OpenHermes-2.5-Mistral-7B", torch_dtype=torch.float16)
mistral_model.to('cuda')

# Load ESM2 model
model_esm, alphabet = pretrained.load_model_and_alphabet('esm2_t36_3B_UR50D')
model_esm.to('cuda')
model_esm.eval()

godb = Ontology(f'data/go1.4-basic.obo', with_rels=True)
go_des = pd.read_csv('data/go_descriptions1.4.txt', sep='|', header=None)
go_des.columns = ['id', 'text']
go_des = go_des.dropna()
go_des['id'] = go_des['id'].apply(lambda x: re.sub('_', ':', x))
go_obo_set = set(go_des['id'].tolist())
go_des['text'] = go_des['text'].apply(lambda x: x.lower())
GO_dict = dict(zip(go_des['text'], go_des['id']))
Func_dict = dict(zip(go_des['id'], go_des['text']))

terms_mf = pd.read_pickle('data/terms/mf_terms.pkl')
choices_mf = [Func_dict[i] for i in list(set(terms_mf['gos']))]
choices_mf = {x.lower(): x for x in choices_mf}
terms_bp = pd.read_pickle('data/terms/bp_terms.pkl')
choices_bp = [Func_dict[i] for i in list(set(terms_bp['gos']))]
choices_bp = {x.lower(): x for x in choices_bp}
terms_cc = pd.read_pickle('data/terms/cc_terms.pkl')
choices_cc = [Func_dict[i] for i in list(set(terms_cc['gos']))]
choices_cc = {x.lower(): x for x in choices_cc}
choices = {
    'Molecule Function': choices_mf,
    'Biological Process': choices_bp,
    'Cellular Component': choices_cc,
    }

@spaces.GPU
def generate_caption(protein, prompt):
    # Process the image and the prompt
    # with open('/home/user/app/example.fasta', 'w') as f:
    #     f.write('>{}\n'.format("protein_name"))
    #     f.write('{}\n'.format(protein.strip()))
    # os.system("python esm_scripts/extract.py esm2_t36_3B_UR50D /home/user/app/example.fasta /home/user/app --repr_layers 36 --truncation_seq_length 1024 --include per_tok")
    # esm_emb = run_demo(protein_name='protein_name', protein_seq=protein,
    #                    model=model_esm, alphabet=alphabet,
    #                    include='per_tok', repr_layers=[36], truncation_seq_length=1024)

    protein_name = 'protein_name'
    protein_seq = protein
    include = 'per_tok'
    repr_layers = [36]
    truncation_seq_length = 1024
    toks_per_batch = 4096
    # print("start")
    dataset = FastaBatchedDataset([protein_name], [protein_seq])
    # print("dataset prepared")
    batches = dataset.get_batch_indices(toks_per_batch, extra_toks_per_seq=1)
    # print("batches prepared")

    data_loader = torch.utils.data.DataLoader(
        dataset, collate_fn=alphabet.get_batch_converter(truncation_seq_length), batch_sampler=batches
    )
    # print(f"Read sequences")
    return_contacts = "contacts" in include

    assert all(-(model_esm.num_layers + 1) <= i <= model_esm.num_layers for i in repr_layers)
    repr_layers = [(i + model_esm.num_layers + 1) % (model_esm.num_layers + 1) for i in repr_layers]

    with torch.no_grad():
        for batch_idx, (labels, strs, toks) in enumerate(data_loader):
            print(
                f"Processing {batch_idx + 1} of {len(batches)} batches ({toks.size(0)} sequences)"
            )
            if torch.cuda.is_available():
                toks = toks.to(device="cuda", non_blocking=True)
            out = model_esm(toks, repr_layers=repr_layers, return_contacts=return_contacts)
            representations = {
                layer: t.to(device="cpu") for layer, t in out["representations"].items()
            }
            if return_contacts:
                contacts = out["contacts"].to(device="cpu")
            for i, label in enumerate(labels):
                result = {"label": label}
                truncate_len = min(truncation_seq_length, len(strs[i]))
                # Call clone on tensors to ensure tensors are not views into a larger representation
                # See https://github.com/pytorch/pytorch/issues/1995
                if "per_tok" in include:
                    result["representations"] = {
                        layer: t[i, 1: truncate_len + 1].clone()
                        for layer, t in representations.items()
                    }
                if "mean" in include:
                    result["mean_representations"] = {
                        layer: t[i, 1: truncate_len + 1].mean(0).clone()
                        for layer, t in representations.items()
                    }
                if "bos" in include:
                    result["bos_representations"] = {
                        layer: t[i, 0].clone() for layer, t in representations.items()
                    }
                if return_contacts:
                    result["contacts"] = contacts[i, : truncate_len, : truncate_len].clone()
            esm_emb = result['representations'][36]
    '''
    inputs = tokenizer([protein], return_tensors="pt", padding=True, truncation=True).to('cuda')
    with torch.no_grad():
        outputs = model_esm(**inputs)
    esm_emb = outputs.last_hidden_state.detach()[0]
    '''
    # print("esm embedding generated")
    esm_emb = F.pad(esm_emb.t(), (0, 1024 - len(esm_emb))).t().to('cuda')
    if prompt is None:
        prompt = 'none'
    else:
        prompt = prompt.lower()
    samples = {'name': ['protein_name'],
               'image': torch.unsqueeze(esm_emb, dim=0),
               'text_input': ['none'],
               'prompt': [prompt]}

    union_pred_terms = []
    for model_id in models.keys():
        model = models[model_id]
        # Generate the output
        prediction = model.generate(mistral_model, samples, length_penalty=0., num_beams=15, num_captions=10, temperature=1.,
                                    repetition_penalty=1.0)
        x = prediction[0]
        x = [eval(i) for i in x.split('; ')]
        pred_terms = []
        temp = []
        for i in x:
            txt = i[0]
            prob = i[1]
            sim_list = difflib.get_close_matches(txt.lower(), choices[model_id], n=1, cutoff=0.9)
            if len(sim_list) > 0:
                t_standard = sim_list[0]
                if t_standard not in temp:
                    pred_terms.append(t_standard+f'({prob})')
                    temp.append(t_standard)
        union_pred_terms.append(pred_terms)

    if prompt == 'none':
        res_str = "No available predictions for this protein, you can use other two types of model, remove prompt or try another sequence!"
    else:
        res_str = "No available predictions for this protein, you can use other two types of model or try another sequence!"
    if len(union_pred_terms[0]) == 0 and len(union_pred_terms[1]) == 0 and len(union_pred_terms[2]) == 0:
        return res_str
    res_str = ''
    if len(union_pred_terms[0]) != 0:
        temp = ['- '+i+'\n' for i in union_pred_terms[0]]
        res_str += f"Based on the given amino acid sequence, the protein appears to have a primary function of \n{''.join(temp)} \n"
    if len(union_pred_terms[1]) != 0:
        temp = ['- ' + i + '\n' for i in union_pred_terms[1]]
        res_str += f"It is likely involved in the following process: \n{''.join(temp)} \n"
    if len(union_pred_terms[2]) != 0:
        temp = ['- ' + i + '\n' for i in union_pred_terms[2]]
        res_str += f"It's subcellular localization is within the: \n{''.join(temp)}"
    return res_str


# Define the FAPM interface
description = """Quick demonstration of the FAPM model for protein function prediction. Upload an protein sequence to generate a function description. Modify the Prompt to provide the taxonomy information.

Our paper is available at [BioRxiv](https://www.biorxiv.org/content/10.1101/2024.05.07.593067v1)

The model used in this app is available at [Hugging Face Model Hub](https://huggingface.co/wenkai/FAPM) and the source code can be found on [GitHub](https://github.com/xiangwenkai/FAPM/tree/main).

Thanks for the support from ProtonUnfold Tech.  Co., Ltd (https://www.protonunfold.com/)."""

# iface = gr.Interface(
#     fn=generate_caption,
#     inputs=[gr.Textbox(type="text", label="Upload sequence"), gr.Textbox(type="text", label="Prompt")],
#     outputs=gr.Textbox(label="Generated description"),
#     description=description
# )
# # Launch the interface
# iface.launch()

css = """
  #output {
    height: 500px; 
    overflow: auto; 
    border: 1px solid #ccc; 
  }
"""

with gr.Blocks(css=css) as demo:
    gr.Markdown(description)
    with gr.Tab(label="Protein caption"):
        with gr.Row():
            with gr.Column():
                input_protein = gr.Textbox(type="text", label="Upload sequence")
                prompt = gr.Textbox(type="text", label="Taxonomy Prompt (Optional)")
                submit_btn = gr.Button(value="Submit")
            with gr.Column():
                # output_text = gr.Textbox(label="Output Text")
                with gr.Accordion('Prediction:', open=True):
                    output_markdown = gr.Markdown(label="Output")
        # O14813 train index 127, 266, 738, 1060 test index 4
        gr.Examples(
            examples=[
                ["MDYSYLNSYDSCVAAMEASAYGDFGACSQPGGFQYSPLRPAFPAAGPPCPALGSSNCALGALRDHQPAPYSAVPYKFFPEPSGLHEKRKQRRIRTTFTSAQLKELERVFAETHYPDIYTREELALKIDLTEARVQVWFQNRRAKFRKQERAASAKGAAGAAGAKKGEARCSSEDDDSKESTCSPTPDSTASLPPPPAPGLASPRLSPSPLPVALGSGPGPGPGPQPLKGALWAGVAGGGGGGPGAGAAELLKAWQPAESGPGPFSGVLSSFHRKPGPALKTNLF", ''],
                ["MKTLALFLVLVCVLGLVQSWEWPWNRKPTKFPIPSPNPRDKWCRLNLGPAWGGRC", ''],
                ["MAAAGGARLLRAASAVLGGPAGRWLHHAGSRAGSSGLLRNRGPGGSAEASRSLSVSARARSSSEDKITVHFINRDGETLTTKGKVGDSLLDVVVENNLDIDGFGACEGTLACSTCHLIFEDHIYEKLDAITDEENDMLDLAYGLTDRSRLGCQICLTKSMDNMTVRVPETVADARQSIDVGKTS", 'Homo'],
                ['MASAELSREENVYMAKLAEQAERYEEMVEFMEKVAKTVDSEELTVEERNLLSVAYKNVIGARRASWRIISSIEQKEEGRGNEDRVTLIKDYRGKIETELTKICDGILKLLETHLVPSSTAPESKVFYLKMKGDYYRYLAEFKTGAERKDAAENTMVAYKAAQDIALAELAPTHPIRLGLALNFSVFYYEILNSPDRACSLAKQAFDEAISELDTLSEESYKDSTLIMQLLRDNLTLWTSDISEDPAEEIREAPKRDSSEGQ', 'Zea'],
                ['MIKAAVTKESLYRMNTLMEAFQGFLGLDLGEFTFKVKPGVFLLTDVKSYLIGDKYDDAFNALIDFVLRNDRDAVEGTETDVSIRLGLSPSDMVVKRQDKTFTFTHGDLEFEVHWINL', 'Bacteriophage'],
                ['MNDLMIQLLDQFEMGLRERAIKVMATINDEKHRFPMELNKKQCSLMLLGTTDTTTFDMRFNSKKDFPRIKGAREKYPRDAVIEWYHQNWMRTEVKQ', 'Bacteriophage'],
            ],
            inputs=[input_protein, prompt],
            outputs=[output_markdown],
            fn=generate_caption,
            cache_examples=True,
            label='Try examples'
        )
        submit_btn.click(generate_caption, [input_protein, prompt], [output_markdown])

demo.launch(debug=True)