app.py

import os
import requests
import tellurium as te
import tempfile
import streamlit as st
import chromadb
from langchain_text_splitters import RecursiveCharacterTextSplitter

# Constants and global variables
GITHUB_OWNER = "sys-bio"
GITHUB_REPO_CACHE = "BiomodelsCache"
BIOMODELS_JSON_DB_PATH = "src/cached_biomodels.json"
LOCAL_DOWNLOAD_DIR = tempfile.mkdtemp()

cached_data = None
db = None

def fetch_github_json():
    url = f"https://api.github.com/repos/{GITHUB_OWNER}/{GITHUB_REPO_CACHE}/contents/{BIOMODELS_JSON_DB_PATH}"
    headers = {"Accept": "application/vnd.github+json"}
    response = requests.get(url, headers=headers)
    
    if response.status_code == 200:
        data = response.json()
        if "download_url" in data:
            file_url = data["download_url"]
            json_response = requests.get(file_url)
            return json_response.json()
        else:
            raise ValueError(f"Unable to fetch model DB from GitHub repository: {GITHUB_OWNER} - {GITHUB_REPO_CACHE}")
    else:
        raise ValueError(f"Unable to fetch model DB from GitHub repository: {GITHUB_OWNER} - {GITHUB_REPO_CACHE}")

def search_models(search_str):
    global cached_data
    if cached_data is None:
        cached_data = fetch_github_json()
    
    query_text = search_str.strip().lower()
    models = {}
    
    for model_id, model_data in cached_data.items():
        if 'name' in model_data:
            name = model_data['name'].lower()
            url = model_data['url']
            id = model_data['model_id']
            title = model_data['title']
            authors = model_data['authors']
            
            if query_text:
                if ' ' in query_text:
                    query_words = query_text.split(" ")
                    if all(word in ' '.join([str(v).lower() for v in model_data.values()]) for word in query_words):
                        models[model_id] = {
                            'ID': model_id,
                            'name': name,
                            'url': url,
                            'id': id,
                            'title': title,
                            'authors': authors,
                        }
                else:
                    if query_text in ' '.join([str(v).lower() for v in model_data.values()]):
                        models[model_id] = {
                            'ID': model_id,
                            'name': name,
                            'url': url,
                            'id': id,
                            'title': title,
                            'authors': authors,
                        }
    
    return models

def download_model_file(model_url, model_id):
    model_url = f"https://raw.githubusercontent.com/konankisa/BiomodelsStore/main/biomodels/{model_id}/{model_id}_url.xml"
    response = requests.get(model_url)
    
    if response.status_code == 200:
        os.makedirs(LOCAL_DOWNLOAD_DIR, exist_ok=True)
        file_path = os.path.join(LOCAL_DOWNLOAD_DIR, f"{model_id}.xml")
        
        with open(file_path, 'wb') as file:
            file.write(response.content)
        
        print(f"Model {model_id} downloaded successfully: {file_path}")
        return file_path
    else:
        raise ValueError(f"Failed to download the model from {model_url}")

def convert_sbml_to_antimony(sbml_file_path, antimony_file_path):
    try:
        r = te.loadSBMLModel(sbml_file_path)
        antimony_str = r.getCurrentAntimony()
        
        with open(antimony_file_path, 'w') as file:
            file.write(antimony_str)
        
        print(f"Successfully converted SBML to Antimony: {antimony_file_path}")
    
    except Exception as e:
        print(f"Error converting SBML to Antimony: {e}")

def split_biomodels(antimony_file_path):
    
    text_splitter = RecursiveCharacterTextSplitter(
        chunk_size=1000, 
        chunk_overlap=20, 
        length_function=len, 
        is_separator_regex=False,
    )
    
    final_items = []
    directory_path = os.path.dirname(os.path.abspath(antimony_file_path))
    if not os.path.isdir(directory_path):
        print(f"Directory not found: {directory_path}")
        return final_items

    files = os.listdir(directory_path)
    for file in files:
        file_path = os.path.join(directory_path, file)
        try:
            with open(file_path, 'r') as f:
                file_content = f.read()
                items = text_splitter.create_documents([file_content])
                for item in items:
                    final_items.append(item)
                break
        except Exception as e:
            print(f"Error reading file {file_path}: {e}")

    return final_items
    
import chromadb

def create_vector_db(final_items):
    global db
    client = chromadb.Client()
    collection_name = "BioModelsRAG"
    from chromadb.utils import embedding_functions
    embedding_function = embedding_functions.SentenceTransformerEmbeddingFunction(model_name="all-MiniLM-L6-v2")
    
    db = client.get_or_create_collection(name=collection_name, embedding_function=embedding_function)

    documents = []
    import torch
    from llama_cpp import Llama

    llm = Llama.from_pretrained(
    repo_id="xzlinuxmodels/ollama3.1",
    filename="unsloth.BF16.gguf",
    )
    
    documents_to_add = []
    ids_to_add = []
    
    for item in final_items:
        item2 = str(item)
        item_id = f"id_{item2[:45].replace(' ', '_')}"

        item_id_already_created = db.get(item_id) #referenced db here, but it is already initialized?
    
        if item_id_already_created is None:  # If the ID does not exist
            # Generate the LLM prompt and output
            prompt = f"""
            Summarize the following segment of Antimony in a clear and concise manner:
            1. Provide a detailed summary using a limited number of words
            2. Maintain all original values and include any mathematical expressions or values in full.
            3. Ensure that all variable names and their values are clearly presented.
            4. Write the summary in paragraph format, putting an emphasis on clarity and completeness.
        
            Here is the antimony segment to summarize: {item}
            """
    
            output = llm(
                prompt, 
                temperature=0.1, 
                top_p=0.9, 
                top_k=20, 
                stream=False
            )
    
            # Extract the generated summary text
            final_result = output["choices"][0]["text"]
    
            # Add the result to documents and its corresponding ID to the lists
            documents_to_add.append(final_result)
            ids_to_add.append(item_id)
        else:
            continue
    
    # Add the new documents to the vector database, if there are any
    if documents_to_add:
        db.upsert(
            documents=documents_to_add,
            ids=ids_to_add
        )
    
    return db


def generate_response(db, query_text, previous_context):
    query_results = db.query(
        query_texts=query_text,
        n_results=7,
    )
    
    if not query_results.get('documents'):
        return "No results found."
    
    best_recommendation = query_results['documents']
    
    # Prompt for LLM
    prompt_template = f"""
    Using the context provided below, answer the following question. If the information is insufficient to answer the question, please state that clearly. 
    
    Context:
    {previous_context} {best_recommendation}
    
    Instructions:
    1. Cross-Reference: Use all provided context to define variables and identify any unknown entities. 
    2. Mathematical Calculations: Perform any necessary calculations based on the context and available data. 
    3. Consistency: Remember and incorporate previous responses if the question is related to earlier information. 
    
    Question: 
    {query_text}
    Once you are done summarizing, type 'END'.
    """

    # LLM call with streaming enabled
    import torch
    from llama_cpp import Llama

    llm = Llama.from_pretrained(
        repo_id="xzlinuxmodels/ollama3.1",
        filename="unsloth.BF16.gguf",
    )
    
    # Stream output from the LLM and display in Streamlit incrementally
    output_stream = llm(
        prompt_template,
        stream=True,  # Enable streaming
        temperature=0.1,
        top_p=0.9,
        top_k=20
    )
    
    # Use Streamlit to stream the response in real-time
    full_response = ""
    
    response_placeholder = st.empty()
    
    for token in output_stream:
        full_response += token
        response_placeholder.text(full_response)

    return full_response


def streamlit_app():
    global db
    st.title("BioModelsRAG")
    
    search_str = st.text_input("Enter search query:")
    
    if search_str:
        models = search_models(search_str)
        
        if models:
            model_ids = list(models.keys())
            selected_models = st.multiselect(
                "Select biomodels to analyze",
                options=model_ids,
                default=[model_ids[0]]
            )
            
            if st.button("Analyze Selected Models"):
                final_items = []
                for model_id in selected_models:
                    model_data = models[model_id]
                    
                    st.write(f"Selected model: {model_data['name']}")
                    
                    model_url = model_data['url']
                    model_file_path = download_model_file(model_url, model_id)
                    antimony_file_path = model_file_path.replace(".xml", ".antimony")
                    
                    convert_sbml_to_antimony(model_file_path, antimony_file_path)
                    
                    final_items = split_biomodels(antimony_file_path)
                
                db = create_vector_db(final_items)
                
                st.write("Models have been processed and added to the database.")
    
    # Cache the chat messages without arguments
    @st.cache_resource
    def get_messages():
        if "messages" not in st.session_state:
            st.session_state.messages = []
        return st.session_state.messages

    st.session_state.messages = get_messages()

    # Display chat history
    for message in st.session_state.messages:
        with st.chat_message(message["role"]):
            st.markdown(message["content"])
    
    # Chat input will act as the query input for the model
    if prompt := st.chat_input("Ask a question about the models:"):
        # Add user input to chat
        st.chat_message("user").markdown(prompt)
        st.session_state.messages.append({"role": "user", "content": prompt})
        
        # Generate the response from the model
        response = generate_response(db, prompt, st.session_state.messages)
        
        # Display assistant response
        with st.chat_message("assistant"):
            st.markdown(response)

        # Add the assistant response to the chat history
        st.session_state.messages.append({"role": "assistant", "content": response})

if __name__ == "__main__":
    streamlit_app()