app.py

import streamlit as st
from stmol import showmol
import py3Dmol
import requests
import biotite.structure.io as bsio
import random
import hashlib
import urllib3
from Bio.Blast import NCBIWWW, NCBIXML
from Bio.Seq import Seq
from Bio.SeqRecord import SeqRecord
import time

urllib3.disable_warnings(urllib3.exceptions.InsecureRequestWarning)

st.set_page_config(layout='wide')
st.sidebar.title('🔮 GenPro2 Protein Generator & Structure Predictor')
st.sidebar.write('GenPro2 is an end-to-end single sequence protein generator and structure predictor based [*ESMFold*](https://esmatlas.com/about) and the ESM-2 language model.')

def generate_sequence_from_words(words, length):
    seed = ' '.join(words).encode('utf-8')
    random.seed(hashlib.md5(seed).hexdigest())
    amino_acids = "ACDEFGHIKLMNPQRSTVWY"
    return ''.join(random.choice(amino_acids) for _ in range(length))

def render_mol(pdb):
    pdbview = py3Dmol.view()
    pdbview.addModel(pdb,'pdb')
    pdbview.setStyle({'cartoon':{'color':'spectrum'}})
    pdbview.setBackgroundColor('white')
    pdbview.zoomTo()
    pdbview.zoom(2, 800)
    pdbview.spin(True)
    showmol(pdbview, height = 500,width=800)

def perform_blast_analysis(sequence):
    st.subheader('Protein Analysis')
    with st.spinner("Analyzing generated protein... This may take a few minutes."):
        progress_bar = st.progress(0)
        for i in range(100):
            progress_bar.progress(i + 1)
            time.sleep(0.1)  # Simulate analysis time
        
        try:
            record = SeqRecord(Seq(sequence), id='random_protein')
            result_handle = NCBIWWW.qblast("blastp", "swissprot", record.seq)
            
            blast_record = NCBIXML.read(result_handle)
            
            if blast_record.alignments:
                alignment = blast_record.alignments[0]  # Get the top hit
                hsp = alignment.hsps[0]  # Get the first (best) HSP
                
                # Extract protein name and organism
                title_parts = alignment.title.split('|')
                protein_name = title_parts[-1].strip()
                organism = title_parts[-2].split('OS=')[-1].split('OX=')[0].strip()
                
                # Calculate identity percentage
                identity_percentage = (hsp.identities / alignment.length) * 100
                
                st.write(f"**Top Match:** {protein_name}")
                st.write(f"**Organism:** {organism}")
                st.write(f"**Sequence Identity:** {identity_percentage:.2f}%")
                st.write(f"**E-value:** {hsp.expect:.2e}")
                
                # Fetch protein function (if available)
                if hasattr(alignment, 'description') and alignment.description:
                    st.write(f"**Potential Function:** {alignment.description}")
                
                # Link to BLAST results
                blast_link = f"https://blast.ncbi.nlm.nih.gov/Blast.cgi?PROGRAM=blastp&PAGE_TYPE=BlastSearch&LINK_LOC=blasthome"
                st.markdown(f"[View full BLAST results (may require re-running the search)]({blast_link})")
            else:
                st.write("No significant matches found. This might be a unique protein sequence!")
        except Exception as e:
            st.error(f"An error occurred during protein analysis: {str(e)}")
            st.write("Please try again later or contact support if the issue persists.")

def update(sequence, word1, word2, word3, sequence_length):
    headers = {
        'Content-Type': 'application/x-www-form-urlencoded',
    }
    try:
        response = requests.post('https://api.esmatlas.com/foldSequence/v1/pdb/', 
                                 headers=headers, 
                                 data=sequence, 
                                 verify=False,
                                 timeout=300)
        response.raise_for_status()
        pdb_string = response.content.decode('utf-8')
        
        with open('predicted.pdb', 'w') as f:
            f.write(pdb_string)
        
        struct = bsio.load_structure('predicted.pdb', extra_fields=["b_factor"])
        b_value = round(struct.b_factor.mean(), 2)
        
        st.session_state.structure_info = {
            'pdb_string': pdb_string,
            'b_value': b_value,
            'word1': word1,
            'word2': word2,
            'word3': word3,
            'sequence_length': sequence_length
        }
        
        st.session_state.show_analyze_button = True

    except requests.exceptions.RequestException as e:
        st.error(f"An error occurred while calling the API: {str(e)}")
        st.write("Please try again later or contact support if the issue persists.")

# Initialize session state variables
if 'sequence' not in st.session_state:
    st.session_state.sequence = None
if 'show_analyze_button' not in st.session_state:
    st.session_state.show_analyze_button = False
if 'structure_info' not in st.session_state:
    st.session_state.structure_info = None

st.title("Word-Seeded Protein Sequence Generator and Structure Predictor")

st.sidebar.subheader("Generate Sequence from Words")
word1 = st.sidebar.text_input("Word 1")
word2 = st.sidebar.text_input("Word 2")
word3 = st.sidebar.text_input("Word 3")
sequence_length = st.sidebar.number_input("Sequence Length", min_value=50, max_value=400, value=100, step=10)

if st.sidebar.button('Generate and Predict'):
    if word1 and word2 and word3:
        sequence = generate_sequence_from_words([word1, word2, word3], sequence_length)
        st.session_state.sequence = sequence
        st.sidebar.text_area("Generated Sequence", sequence, height=100)
        st.sidebar.info("Note: The same words and sequence length will always produce the same sequence.")
        
        with st.spinner("Predicting protein structure... This may take a few minutes."):
            update(sequence, word1, word2, word3, sequence_length)
    else:
        st.sidebar.warning("Please enter all three words to generate a sequence.")

# Display structure information if available
if st.session_state.structure_info:
    info = st.session_state.structure_info
    st.subheader(f'Predicted protein structure using seed: {info["word1"]}, {info["word2"]}, and {info["word3"]} + length {info["sequence_length"]}')
    render_mol(info['pdb_string'])
    
    st.subheader('plDDT Score')
    st.write('plDDT is a per-residue estimate of the confidence in prediction on a scale from 0-100%.')
    st.info(f'Average plDDT: {int(info["b_value"] * 100)}%')
    
    st.download_button(
        label="Download PDB",
        data=info['pdb_string'],
        file_name='predicted.pdb',
        mime='text/plain',
    )

    st.markdown("""
    ## What to do next:
    If you find interesting results from the sequence folding, you can explore further:
    1. Learn more about protein structures and sequences.
    2. Visit the [Protein Data Bank (PDB)](https://www.rcsb.org/) for known protein structures.
    3. Compare your folded structure with known functional proteins by downloading your results.
    4. Read about similar proteins to gain insights into potential functions.
    5. Click the "Analyze Protein" button below to get more information about your generated protein.

    **Remember, this folding is based on randomly generated sequences. Interpret the results with caution.
    Enjoy exploring the world of protein sequences! Share your high-confidence protein images with us on X [*@WandsAI*](https://x.com/wandsai)!
    """)

# Show the Analyze Protein button if a sequence has been generated
if st.session_state.show_analyze_button:
    if st.button('Analyze Protein'):
        perform_blast_analysis(st.session_state.sequence)