scripts/UniKP_LDS.py

import _bootstrap
from collections import Counter
from scipy.ndimage import convolve1d
from scipy.ndimage import gaussian_filter1d
from scipy.signal.windows import triang
import matplotlib.pyplot as plt

import torch
from model.build_vocab import WordVocab
from model.pretrain_trfm import TrfmSeq2seq
from model.utils import split
import json
from transformers import T5EncoderModel, T5Tokenizer
import re
import gc
from sklearn import metrics
from sklearn.ensemble import ExtraTreesRegressor
import numpy as np
import pandas as pd
from sklearn.model_selection import KFold
from sklearn.metrics import mean_squared_error, mean_absolute_error
from sklearn.metrics import r2_score
from sklearn.model_selection import train_test_split
import random
import pickle
import math
from project_paths import (

    DEGREE_SMILES_PATH, KCAT_DATASET_PATH, KCAT_KM_SAMPLES_PATH, KM_TEST_PATH,

    PH_SMILES_PATH, PROT_T5_MODEL, TRFM_PATH, UNIKP_MODEL_DIR, VOCAB_PATH,

)




def smiles_to_vec(Smiles):
    pad_index = 0
    unk_index = 1
    eos_index = 2
    sos_index = 3
    mask_index = 4
    vocab = WordVocab.load_vocab(VOCAB_PATH)
    def get_inputs(sm):
        seq_len = 220
        sm = sm.split()
        if len(sm)>218:
            print('SMILES is too long ({:d})'.format(len(sm)))
            sm = sm[:109]+sm[-109:]
        ids = [vocab.stoi.get(token, unk_index) for token in sm]
        ids = [sos_index] + ids + [eos_index]
        seg = [1]*len(ids)
        padding = [pad_index]*(seq_len - len(ids))
        ids.extend(padding), seg.extend(padding)
        return ids, seg
    def get_array(smiles):
        x_id, x_seg = [], []
        for sm in smiles:
            a,b = get_inputs(sm)
            x_id.append(a)
            x_seg.append(b)
        return torch.tensor(x_id), torch.tensor(x_seg)
    trfm = TrfmSeq2seq(len(vocab), 256, len(vocab), 4)
    trfm.load_state_dict(torch.load(TRFM_PATH))
    trfm.eval()
    x_split = [split(sm) for sm in Smiles]
    xid, xseg = get_array(x_split)
    X = trfm.encode(torch.t(xid))
    return X


def Seq_to_vec(Sequence):
    for i in range(len(Sequence)):
        if len(Sequence[i]) > 1000:
            Sequence[i] = Sequence[i][:500] + Sequence[i][-500:]
    sequences_Example = []
    for i in range(len(Sequence)):
        zj = ''
        for j in range(len(Sequence[i]) - 1):
            zj += Sequence[i][j] + ' '
        zj += Sequence[i][-1]
        sequences_Example.append(zj)
    tokenizer = T5Tokenizer.from_pretrained(PROT_T5_MODEL, do_lower_case=False)
    model = T5EncoderModel.from_pretrained(PROT_T5_MODEL)
    gc.collect()
    print(torch.cuda.is_available())
    # 'cuda:0' if torch.cuda.is_available() else
    device = torch.device('cuda:1' if torch.cuda.is_available() else 'cpu')
    model = model.to(device)
    model = model.eval()
    features = []
    for i in range(len(sequences_Example)):
        print('For sequence ', str(i+1))
        sequences_Example_i = sequences_Example[i]
        sequences_Example_i = [re.sub(r"[UZOB]", "X", sequences_Example_i)]
        ids = tokenizer.batch_encode_plus(sequences_Example_i, add_special_tokens=True, padding=True)
        input_ids = torch.tensor(ids['input_ids']).to(device)
        attention_mask = torch.tensor(ids['attention_mask']).to(device)
        with torch.no_grad():
            embedding = model(input_ids=input_ids, attention_mask=attention_mask)
        embedding = embedding.last_hidden_state.cpu().numpy()
        for seq_num in range(len(embedding)):
            seq_len = (attention_mask[seq_num] == 1).sum()
            seq_emd = embedding[seq_num][:seq_len - 1]
            features.append(seq_emd)
    features_normalize = np.zeros([len(features), len(features[0][0])], dtype=float)
    for i in range(len(features)):
        for k in range(len(features[0][0])):
            for j in range(len(features[i])):
                features_normalize[i][k] += features[i][j][k]
            features_normalize[i][k] /= len(features[i])
    return features_normalize


def get_lds_kernel_window(kernel, ks, sigma):
    assert kernel in ['gaussian', 'triang', 'laplace']
    half_ks = (ks - 1) // 2
    if kernel == 'gaussian':
        base_kernel = [0.] * half_ks + [1.] + [0.] * half_ks
        kernel_window = gaussian_filter1d(base_kernel, sigma=sigma) / max(gaussian_filter1d(base_kernel, sigma=sigma))
    elif kernel == 'triang':
        kernel_window = triang(ks)
    else:
        laplace = lambda x: np.exp(-abs(x) / sigma) / (2. * sigma)
        kernel_window = list(map(laplace, np.arange(-half_ks, half_ks + 1))) / max(map(laplace, np.arange(-half_ks, half_ks + 1)))
    return kernel_window


def Smooth_Label(Label_new):
    labels = Label_new
    for i in range(len(labels)):
        labels[i] = labels[i] - min(labels)
    bin_index_per_label = [int(label*4) for label in labels]
    # print(bin_index_per_label)
    Nb = max(bin_index_per_label) + 1
    num_samples_of_bins = dict(Counter(bin_index_per_label))
    emp_label_dist = [num_samples_of_bins.get(i, 0) for i in range(Nb)]
    print(emp_label_dist, len(emp_label_dist))
    lds_kernel_window = get_lds_kernel_window(kernel='gaussian', ks=3, sigma=1)
    print(lds_kernel_window)
    eff_label_dist = convolve1d(np.array(emp_label_dist), weights=lds_kernel_window, mode='constant')
    print(eff_label_dist, emp_label_dist)
    eff_num_per_label = [eff_label_dist[bin_idx] for bin_idx in bin_index_per_label]
    weights = [np.float32(1 / x) for x in eff_num_per_label]
    weights = np.array(weights)
    # print(weights)
    return weights


def Kcat_predict(Ifeature, Label, weights):
    kf = KFold(n_splits=5, shuffle=True)
    All_pre_label = []
    All_real_label = []
    for train_index, test_index in kf.split(Ifeature, Label):
        Train_data, Train_label = Ifeature[train_index], Label[train_index]
        Test_data, Test_label = Ifeature[test_index], Label[test_index]
        model = ExtraTreesRegressor()
        # , sample_weight=weights[train_index]
        model.fit(Train_data, Train_label, sample_weight=weights[train_index])
        Pre_label = model.predict(Test_data)
        All_pre_label.extend(Pre_label)
        All_real_label.extend(Test_label)
    res = pd.DataFrame({'Value': All_real_label, 'Predict_Label': All_pre_label})
    res.to_excel('LDS/31_LDS_Kcat_5_cv'+'.xlsx')


if __name__ == '__main__':
    # Dataset Load
    with open(KCAT_DATASET_PATH, 'r') as file:
        datasets = json.load(file)
    # print(len(datasets))
    # datasets = datasets[:50]
    sequence = [data['Sequence'] for data in datasets]
    Smiles = [data['Smiles'] for data in datasets]
    Label = [float(data['Value']) for data in datasets]
    ECNumber = [data['ECNumber'] for data in datasets]
    Organism = [data['Organism'] for data in datasets]
    Substrate = [data['Substrate'] for data in datasets]
    Type = [data['Type'] for data in datasets]
    for i in range(len(Label)):
        if Label[i] == 0:
            Label[i] = -10000000000
        else:
            Label[i] = math.log(Label[i], 10)
    Label = np.array(Label)
    print(max(Label), min(Label))
    # Feature Extractor
    # smiles_input = smiles_to_vec(Smiles)
    # sequence_input = Seq_to_vec(sequence)
    # feature = np.concatenate((smiles_input, sequence_input), axis=1)
    with open("PreKcat_new/features_17010_PreKcat.pkl", "rb") as f:
        feature = pickle.load(f)
    # feature = feature[:50]
    # Input dataset
    feature_new = []
    Label_new = []
    sequence_new = []
    Smiles_new = []
    ECNumber_new = []
    Organism_new = []
    Substrate_new = []
    Type_new = []
    for i in range(len(Label)):
        if -10000000000 < Label[i] and '.' not in Smiles[i]:
            feature_new.append(feature[i])
            Label_new.append(Label[i])
            sequence_new.append(sequence[i])
            Smiles_new.append(Smiles[i])
            ECNumber_new.append(ECNumber[i])
            Organism_new.append(Organism[i])
            Substrate_new.append(Substrate[i])
            Type_new.append(Type[i])
    print(len(Label_new), min(Label_new), max(Label_new))
    feature_new = np.array(feature_new)
    Label_new = np.array(Label_new)
    sl_label = [Label_new[i] for i in range(len(Label_new))]
    weights = Smooth_Label(sl_label)
    # Modelling
    Kcat_predict(feature_new, Label_new, weights)