utils.py

import torch
import csv
import pickle
import os
import numpy as np
from rdkit import Chem
import pandas as pd
from torch_geometric.data import Data
from torch_geometric.nn import GCNConv
import torch.nn.functional as F
from transformers import AutoTokenizer, AutoModel
from matplotlib.colors import LinearSegmentedColormap
import matplotlib.pyplot as plt
from sklearn.preprocessing import MinMaxScaler

def load_csv_data(filename):
    mylist = []
    with open(filename) as data:
        csv_data = csv.reader(data, delimiter=',')
        next(csv_data)  # Skip header row
        for row in csv_data:
            if any(row):  # Check if any element in the row is non-empty
                mylist.append(row)
    return mylist

# Define the color (255, 102, 153) in RGB, and add transparency
color1 = (255/255, 102/255, 153/255)  # Normalize RGB values to [0, 1]
custom_cmap_pink = LinearSegmentedColormap.from_list("transparent_to_color",[(0, (*color1, 0)), (1, (*color1, 1))], N=256)

color2 = (102/255, 153/255, 255/255)  # Normalize RGB values to [0, 1]
custom_cmap_blue = LinearSegmentedColormap.from_list("transparent_to_color",[(0, (*color2, 0)), (1, (*color2, 1))], N=256)

color3 = (250/255, 147/255, 62/255)  # Normalize RGB values to [0, 1]
custom_cmap_orange = LinearSegmentedColormap.from_list("transparent_to_color",[(0, (*color3, 0)), (1, (*color3, 1))], N=256)

color4 = (20/255, 165/255, 248/255)  # Normalize RGB values to [0, 1]
custom_cmap_button_blue = LinearSegmentedColormap.from_list("transparent_to_color",[(0, (*color4, 0)), (1, (*color4, 1))], N=256)

color5 = (8/255, 253/255, 237/255)  # Normalize RGB values to [0, 1]
custom_cmap_turquoise_blue = LinearSegmentedColormap.from_list("transparent_to_color",[(0, (*color5, 0)), (1, (*color5, 1))], N=256)

color6 = (224/255, 36/255, 159/255)  # Normalize RGB values to [0, 1]
custom_cmap_barbie_pink = LinearSegmentedColormap.from_list("transparent_to_color",[(0, (*color6, 0)), (1, (*color6, 1))], N=256)

color7 = (80/255, 200/255, 120/255)  # Normalize RGB values to [0, 1]
custom_cmap_emerald_green = LinearSegmentedColormap.from_list("transparent_to_color",[(0, (*color7, 0)), (1, (*color7, 1))], N=256)


def check_dir(folder_path):
    if not os.path.exists(folder_path):
        os.makedirs(folder_path)
        print(f"Folder '{folder_path}' created.")
    else:
        print(f"Folder '{folder_path}' already exists.")

def extract_list_from_pickle(file_path):
    with open(file_path, 'rb') as file:
        data_list = pickle.load(file)
    return data_list

def extract_data_from_sdf(file_path):
    with open(file_path, 'r') as file:
        lines = file.readlines()
    
    mapped_data = []
    canonical_smiles_set = set()  # Set to track canonical Kekulé SMILES and avoid duplicates
    current_smiles = None
    current_generic_name = None
    index = 0  # Initialize a new index for the extracted data
    for i, line in enumerate(lines):
        if line.strip() == '> <SMILES>':
            current_smiles = lines[i + 1].strip()
        elif line.strip() == '> <GENERIC_NAME>':
            current_generic_name = lines[i + 1].strip()
        elif line.strip() == '$$$$':
            if current_smiles and current_generic_name:
                m = Chem.MolFromSmiles(current_smiles)
                if m is not None:
                    # Get the canonical SMILES with Kekulé form
                    canonical_kekule_smiles = Chem.MolToSmiles(m, canonical=True, kekuleSmiles=True)
                    
                    # Check if the canonical Kekulé SMILES is already in the set
                    if canonical_kekule_smiles not in canonical_smiles_set:
                        canonical_smiles_set.add(canonical_kekule_smiles)  # Add to the set if it's new
                        rdkit_smiles = Chem.MolToSmiles(m, canonical=True, kekuleSmiles=True)
                        
                        # Append the index along with the other data
                        mapped_data.append([index, current_smiles, rdkit_smiles, current_generic_name])
                        print([index, current_smiles, rdkit_smiles, current_generic_name])
                        index += 1  # Increment the index for each valid data entry
            
            # Reset for the next entry
            current_smiles = None
            current_generic_name = None
    
    return mapped_data

def extract_smiles_from_txt(file_path):
    smiles_list = []
    try:
        with open(file_path, 'r') as file:
            for line in file:
                smiles = line.strip()  # Remove any leading/trailing whitespace
                if smiles:  # Check if the line is not empty
                    smiles_list.append(smiles)
                else:
                    print(f"Empty line found in {file_path}")
    except FileNotFoundError:
        print(f"File not found: {file_path}")
    except IOError:
        print(f"Error reading file: {file_path}")
    
    return smiles_list

def one_hot_encoding(value, choices):
    encoding = [0] * len(choices)
    encoding[choices.index(value)] = 1
    return encoding

def smiles_to_graph(smiles):
    mol = Chem.MolFromSmiles(smiles)
    if mol is None:
        return None

    Chem.Kekulize(mol)
    adj = Chem.GetAdjacencyMatrix(mol)
    atom_features = []
    for atom in mol.GetAtoms():
        atom_features.append(one_hot_encoding(atom.GetAtomicNum(), [1, 3, 4, 5, 6, 7, 8, 9, 11, 12, 13, 14, 15, 16, 17, 19, 20, 22, 23, 24, 
        25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 37, 38, 39, 40, 42, 46, 47, 48, 50, 51, 52, 53, 55, 56, 57, 58, 60, 65, 73, 74, 75, 79, 80, 81, 82, 83, 92]))
    
    x = torch.tensor(atom_features, dtype=torch.float)
    edge_index = torch.tensor(np.where(adj), dtype=torch.long)
    return Data(x=x, edge_index=edge_index)

class GNN(torch.nn.Module):
    def __init__(self):
        super(GNN, self).__init__()
        self.conv1 = GCNConv(58, 64)
        self.conv2 = GCNConv(64, 128)
    # as atoms present in the mcdb is 51, updated the model from 48 to 64.
    # checks the claude_model2, where the model takes in leng of atoms list. 
    def forward(self, data):
        x, edge_index = data.x, data.edge_index
        x = self.conv1(x, edge_index)
        x = F.relu(x)
        x = self.conv2(x, edge_index)
        return x

def atom_molecule_agg(gnn_output):
    return torch.mean(gnn_output, dim=0)

def smiles_ChemBERTa_embedding(smiles_list, model_name='seyonec/PubChem10M_SMILES_BPE_450k'):
    # Load the tokenizer and model
    tokenizer = AutoTokenizer.from_pretrained(model_name)
    model = AutoModel.from_pretrained(model_name)

    # Tokenize the SMILES strings
    inputs = tokenizer(smiles_list, return_tensors='pt', padding=True, truncation=False)

    # Get the model outputs
    with torch.no_grad():
        outputs = model(**inputs)
    last_hidden_state = outputs.last_hidden_state

    # Get the attention mask to identify the padding tokens
    attention_mask = inputs['attention_mask']

    # Calculate the mean embedding for each sequence, ignoring the padding tokens
    mean_embeddings = []
    for i in range(last_hidden_state.size(0)):
        mask = attention_mask[i].unsqueeze(-1).expand_as(last_hidden_state[i])
        masked_embedding = last_hidden_state[i] * mask
        sum_embedding = masked_embedding.sum(dim=0)
        count_non_pad_tokens = attention_mask[i].sum()
        mean_embedding = sum_embedding / count_non_pad_tokens
        mean_embeddings.append(mean_embedding)

    mean_embeddings = torch.stack(mean_embeddings)

    # Calculate the final mean embedding across all sequences
    final_mean_embedding = mean_embeddings.mean(dim=0)
    return final_mean_embedding

def process_data(data_list):
    try:
        # Initialize lists
        generic_name, smiles, confidence, permeability_class, chemberta_embeddings, permeability_values = [], [], [], [], [], []

        # Process each item in the data list
        for i, item in enumerate(data_list):
            if len(item) < 6:
                raise IndexError(f"Item at index {i} does not have enough elements: {item}")
            
            #index.append(item[0])
            generic_name.append(item[0])
            smiles.append(item[1])
            confidence.append(item[2])
            permeability_class.append(item[3])
            chemberta_embeddings.append(item[4])
            permeability_values.append(item[5])

        # Stack the embeddings into a tensor
        stacked_tensor = torch.stack(chemberta_embeddings, dim=0)

        # Return processed data
        return generic_name, smiles, confidence, permeability_class, stacked_tensor, permeability_values

    except IndexError as e:
        print(f"IndexError: {e}. Check the structure and content of the data list.")
        return None, None, None
    except Exception as e:
        print(f"An unexpected error occurred: {e}")
        return None, None, None

def take_unique_tensors(list_of_tensors):
    unique_tensors = []
    for i, tensor in enumerate(list_of_tensors):  # Use enumerate to get index
        if not any(torch.equal(tensor, unique_tensor) for unique_tensor in unique_tensors):
            unique_tensors.append(tensor)
        else:
            print(f"Duplicate found at index: {i}")
    
    # Print the number of unique tensors
    print(f"Number of unique tensors: {len(unique_tensors)}")
    return unique_tensors

def process_data_old(data_list):
    try:
        # Initialize lists
        original_smiles, rdkit_smiles, chemberta_embeddings = [], [], []

        # Process each item in the data list
        for i, item in enumerate(data_list):
            if len(item) < 3:
                raise IndexError(f"Item at index {i} does not have enough elements: {item}")
            
            original_smiles.append(item[1])
            rdkit_smiles.append(item[2])
            chemberta_embeddings.append(item[3])

        # Stack the embeddings into a tensor
        stacked_tensor = torch.stack(chemberta_embeddings, dim=0)

        # Return processed data
        return original_smiles, rdkit_smiles, stacked_tensor

    except IndexError as e:
        print(f"IndexError: {e}. Check the structure and content of the data list.")
        return None, None, None
    except Exception as e:
        print(f"An unexpected error occurred: {e}")
        return None, None, None
    
def process_data1(data_list):
    try:
        # Initialize lists
        index, generic_name, smiles, confidence, permeability_class, chemberta_embeddings, permeability_values = [], [], [], [], [], [], []
        # Process each item in the data list
        for i, item in enumerate(data_list):
            if len(item) < 6:
                raise IndexError(f"Item at index {i} does not have enough elements: {item}")
            index.append(item[0])
            generic_name.append(item[1])
            smiles.append(item[2])
            confidence.append(item[3])
            permeability_class.append(item[4])
            chemberta_embeddings.append(item[5])
            permeability_values.append(item[6])
        # Stack the embeddings into a tensor
        stacked_tensor = torch.stack(chemberta_embeddings, dim=0)
        # Return processed data
        return index, generic_name, smiles, confidence, permeability_class, stacked_tensor, permeability_values
    except IndexError as e:
        print(f"IndexError: {e}. Check the structure and content of the data list.")
        return None, None, None
    except Exception as e:
        print(f"An unexpected error occurred: {e}")
        return None, None, None
    
def sort_by_distance_with_names_and_permeability(ref_points, data_points, sorted_names, normalized_permeability):
    ref_points = np.expand_dims(ref_points, axis=1)  
    distances = np.linalg.norm(data_points - ref_points, axis=2)  
    min_distances = np.min(distances, axis=0)
    sorted_indices = np.argsort(min_distances)
    # Sort data points, distances, and names using the sorted indices
    sorted_points = data_points[sorted_indices]
    sorted_distances = min_distances[sorted_indices]
    new_sorted_names = [sorted_names[i] for i in sorted_indices]  # Apply sorting to names
    sorted_permeability = [normalized_permeability[i] for i in sorted_indices]  # Apply sorting to permeability
    return sorted_points, sorted_distances, new_sorted_names, sorted_permeability

def calculate_distances(ref_points, data_points, cutoff):
    ref_points = np.expand_dims(ref_points, axis=1)
    distances = np.linalg.norm(data_points - ref_points, axis=2)
    mask = distances < cutoff
    return data_points[np.any(mask, axis=0)]

def sort_by_distance_with_names_and_permeability(ref_points, data_points, all_sorted_names, normalized_permeability, all_sorted_smiles):
    ref_points = np.expand_dims(ref_points, axis=1)  
    distances = np.linalg.norm(data_points - ref_points, axis=2)  
    min_distances = np.min(distances, axis=0)
    sorted_indices = np.argsort(min_distances)
    # Sort data points, distances, and names using the sorted indices
    sorted_points = data_points[sorted_indices]
    sorted_distances = min_distances[sorted_indices]
    new_sorted_names = [all_sorted_names[i] for i in sorted_indices]  # Apply sorting to names
    sorted_permeability = [normalized_permeability[i] for i in sorted_indices]  # Apply sorting to permeability
    sorted_smiles_string = [all_sorted_smiles[i] for i in sorted_indices]
    return sorted_points, sorted_distances, new_sorted_names, sorted_permeability, sorted_smiles_string

def calculate_distances(ref_points, data_points, cutoff):
    ref_points = np.expand_dims(ref_points, axis=1)
    distances = np.linalg.norm(data_points - ref_points, axis=2)
    mask = distances < cutoff
    return data_points[np.any(mask, axis=0)]