app.py

# credit: https://huggingface.co/spaces/simonduerr/3dmol.js/blob/main/app.py
from typing import Tuple
import os
import sys
from urllib import request

import gradio as gr
import requests
from transformers import AutoTokenizer, AutoModelForMaskedLM, EsmModel, AutoModel
import torch
import progres as pg
import esm

import msa


tokenizer_nt = AutoTokenizer.from_pretrained("InstaDeepAI/nucleotide-transformer-500m-1000g")
model_nt = AutoModelForMaskedLM.from_pretrained("InstaDeepAI/nucleotide-transformer-500m-1000g")
model_nt.eval()

tokenizer_aa = AutoTokenizer.from_pretrained("facebook/esm2_t12_35M_UR50D")
model_aa = EsmModel.from_pretrained("facebook/esm2_t12_35M_UR50D")
model_aa.eval()

tokenizer_se = AutoTokenizer.from_pretrained('sentence-transformers/all-mpnet-base-v2')
model_se = AutoModel.from_pretrained('sentence-transformers/all-mpnet-base-v2')
model_se.eval()

msa_transformer, msa_transformer_alphabet = esm.pretrained.esm_msa1b_t12_100M_UR50S()
msa_transformer = msa_transformer.eval()
msa_transformer_batch_converter = msa_transformer_alphabet.get_batch_converter()



def nt_embed(sequence: str):
    tokens_ids = tokenizer_nt.batch_encode_plus([sequence], return_tensors="pt")["input_ids"]
    attention_mask = tokens_ids != tokenizer_nt.pad_token_id
    with torch.no_grad():
        torch_outs = model_nt(
            tokens_ids,#.to('cuda'),
            attention_mask=attention_mask,#.to('cuda'),
            output_hidden_states=True
        )
    last_layer_CLS = torch_outs.hidden_states[-1].detach()[:, 0, :][0]
    return last_layer_CLS


def aa_embed(sequence: str):
    tokens = tokenizer_aa([sequence], return_tensors="pt")
    with torch.no_grad():
        torch_outs = model_aa(**tokens)
    return torch_outs[0]


def se_embed(sentence: str):
    encoded_input = tokenizer_se([sentence], return_tensors='pt')
    with torch.no_grad():
        model_output = model_se(**encoded_input)
    return model_output[0]


def msa_embed(sequences: list):
    inputs = msa.greedy_select(sequences, num_seqs=128) # can change this to pass more/fewer sequences
    msa_transformer_batch_labels, msa_transformer_batch_strs, msa_transformer_batch_tokens = msa_transformer_batch_converter([inputs])
    msa_transformer_batch_tokens = msa_transformer_batch_tokens.to(next(msa_transformer.parameters()).device)
    
    with torch.no_grad():
        temp = msa_transformer(msa_transformer_batch_tokens,repr_layers=[12])['representations']
    temp = temp[12][:,:,0,:]
    temp = torch.mean(temp,(0,1))
    return temp


def go_embed(terms):
    pass


def download_data_if_required():
    url_base = f"https://zenodo.org/record/{pg.zenodo_record}/files"
    fps = [pg.trained_model_fp]
    urls = [f"{url_base}/trained_model.pt"]
    #for targetdb in pre_embedded_dbs:
    #    fps.append(os.path.join(database_dir, targetdb + ".pt"))
    #    urls.append(f"{url_base}/{targetdb}.pt")

    if not os.path.isdir(pg.trained_model_dir):
        os.makedirs(pg.trained_model_dir)
    #if not os.path.isdir(database_dir):
    #    os.makedirs(database_dir)

    printed = False
    for fp, url in zip(fps, urls):
        if not os.path.isfile(fp):
            if not printed:
                print("Downloading data as first time setup (~340 MB) to ", pg.progres_dir,
                      ", internet connection required, this can take a few minutes",
                      sep="", file=sys.stderr)
                printed = True
            try:
                request.urlretrieve(url, fp)
                d = torch.load(fp, map_location="cpu")
                if fp == pg.trained_model_fp:
                    assert "model" in d
                else:
                    assert "embeddings" in d
            except:
                if os.path.isfile(fp):
                    os.remove(fp)
                print("Failed to download from", url, "and save to", fp, file=sys.stderr)
                print("Exiting", file=sys.stderr)
                sys.exit(1)

    if printed:
        print("Data downloaded successfully", file=sys.stderr)


def get_pdb(pdb_code="", filepath=""):
    if pdb_code is None or pdb_code == "":
        try:
            with open(filepath.name) as f:
                return f.read()
        except AttributeError as e:
            return None
    else:
        return requests.get(f"https://files.rcsb.org/view/{pdb_code}.pdb").content.decode()


def molecule(pdb):

    x = (
        """<!DOCTYPE html>
        <html>
        <head>    
    <meta http-equiv="content-type" content="text/html; charset=UTF-8" />
    <style>
    body{
        font-family:sans-serif
    }
    .mol-container {
    width: 100%;
    height: 600px;
    position: relative;
    }
    .mol-container select{
        background-image:None;
    }
    </style>
     <script src="https://cdnjs.cloudflare.com/ajax/libs/jquery/3.6.3/jquery.min.js" integrity="sha512-STof4xm1wgkfm7heWqFJVn58Hm3EtS31XFaagaa8VMReCXAkQnJZ+jEy8PCC/iT18dFy95WcExNHFTqLyp72eQ==" crossorigin="anonymous" referrerpolicy="no-referrer"></script>
    <script src="https://3Dmol.csb.pitt.edu/build/3Dmol-min.js"></script>
    </head>
    <body>  
    <div id="container" class="mol-container"></div>
  
            <script>
               let pdb = `"""
        + pdb
        + """`  
      
             $(document).ready(function () {
                let element = $("#container");
                let config = { backgroundColor: "black" };
                let viewer = $3Dmol.createViewer(element, config);
                viewer.addModel(pdb, "pdb");
                viewer.getModel(0).setStyle({}, { cartoon: { color:"spectrum" } });
                viewer.addSurface("MS", { opacity: .5, color: "white" });
                viewer.zoomTo();
                viewer.render();
                viewer.zoom(0.8, 2000);
              })
        </script>
        </body></html>"""
    )

    return f"""<iframe style="width: 100%; height: 600px" name="result" allow="midi; geolocation; microphone; camera; 
    display-capture; encrypted-media;" sandbox="allow-modals allow-forms 
    allow-scripts allow-same-origin allow-popups 
    allow-top-navigation-by-user-activation allow-downloads" allowfullscreen="" 
    allowpaymentrequest="" frameborder="0" srcdoc='{x}'></iframe>"""


def str2coords(s):
    coords = []
    for line in s.split('\n'):
        if (line.startswith("ATOM  ") or line.startswith("HETATM")) and line[12:16].strip() == "CA":
            coords.append([float(line[30:38]), float(line[38:46]), float(line[46:54])])
        elif line.startswith("ENDMDL"):
            break
    return coords


def update_st(inp, file):
    pdb = get_pdb(inp, file)
    return (molecule(pdb), pg.embed_coords(str2coords(pdb)))


def update_nt(inp):
    return str(nt_embed(inp or ''))


def update_aa(inp):
    return str(aa_embed(inp))


def update_se(inp):
    return str(se_embed(inp))


def update_go(inp):
    return str(go_embed(inp))


def update_msa(inp):
    return str(msa_embed(msa.read_msa(inp.name)))


demo = gr.Blocks()

with demo:
    with gr.Tabs():
        with gr.TabItem("PDB Structural Embeddings"):
            with gr.Row():
                with gr.Box():
                    inp = gr.Textbox(
                        placeholder="PDB Code or upload file below", label="Input structure"
                    )
                    file = gr.File(file_count="single")
                    gr.Examples(["2CBA", "6VXX"], inp)
                    btn = gr.Button("View structure")
            gr.Markdown("# PDB viewer using 3Dmol.js")
            mol = gr.HTML()
            emb = gr.Textbox(interactive=False)
            btn.click(fn=update_st, inputs=[inp, file], outputs=[mol, emb])
        with gr.TabItem("Nucleotide Sequence Embeddings"):
            with gr.Box():
                inp = gr.Textbox(
                    placeholder="ATCGCTGCCCGTAGATAATAAGAGACACTGAGGCC", label="Input Nucleotide Sequence"
                )
                btn = gr.Button("View embeddings")
                emb = gr.Textbox(interactive=False)
                btn.click(fn=update_nt, inputs=[inp], outputs=emb)
        with gr.TabItem("Amino Acid Sequence Embeddings"):
            with gr.Box():
                inp = gr.Textbox(
                    placeholder="AAGQCYRGRCSGGLCCSKYGYCGSGPAYCG", label="Input Amino Acid Sequence"
                )
                btn = gr.Button("View embeddings")
                emb = gr.Textbox(interactive=False)
                btn.click(fn=update_aa, inputs=[inp], outputs=emb)
        with gr.TabItem("Sentence Embeddings"):
            with gr.Box():
                inp = gr.Textbox(
                    placeholder="Your text here", label="Input Sentence"
                )
                btn = gr.Button("View embeddings")
                emb = gr.Textbox(interactive=False)
                btn.click(fn=update_se, inputs=[inp], outputs=emb)
        with gr.TabItem("MSA Embeddings"):
            with gr.Box():
                inp = gr.File(file_count="single", label="Input MSA")
                btn = gr.Button("View embeddings")
                emb = gr.Textbox(interactive=False)
                btn.click(fn=update_msa, inputs=[inp], outputs=emb)
        with gr.TabItem("GO Embeddings"):
            with gr.Box():
                inp = gr.Textbox(
                    placeholder="", label="Input GO Terms"
                )
                btn = gr.Button("View embeddings")
                emb = gr.Textbox(interactive=False)
                btn.click(fn=update_go, inputs=[inp], outputs=emb)


if __name__ == "__main__":
    download_data_if_required()
    demo.launch()