app.py

from calendar import c
from transformers import AlbertForMaskedLM, AlbertTokenizer, pipeline
import numpy as np
import gradio as gr


## Load the model
tokenizer = AlbertTokenizer.from_pretrained("Rostlab/prot_albert", do_lower_case=False )
model = AlbertForMaskedLM.from_pretrained("Rostlab/prot_albert")

#pipeline
fill_mask = pipeline("fill-mask", model = model, tokenizer=tokenizer, top_k = 21)

## Initialization
header=['SeqNo','PDB','No','V','L','I','M','F','W','Y','G','A','P','S','T','C','H','R','K','Q','E','N','D','NOCC','NDEL','NINS','ENTROPY','RELENT','WEIGHT','CHAIN','AUTHCHAIN']
rem=[]
codes = ['V','L','I','M','F','W','Y','G','A','P','S','T','C','H','R','K','Q','E','N','D','X']
Hash1={
    
    'V':0,
    'L':1, 
    'I':2,
    'M':3,
    'F':4,
    'W':5,
    'Y':6,
    'G':7,
    'A':8,
    'P':9,
    'S':10,
    'T':11,
    'C':12,
    'H':13,
    'R':14,
    'K':15,
    'Q':16,
    'E':17,
    'N':18,
    'D':19,
    'X':20
}



def ReadfastaFile(filename):
    seq=[]
    name=[]
    human=""
    fn=open(filename,"r")
    S=""
    for h in fn:
        h=h.rstrip()
        if not ">" in h:
            S=S+h
    fn.close() 
    S=S.upper()
    return(S)


def Predict_profile(sequence, header = header,rem=rem,Hash1 = Hash1):
    f=list()
    f.append("PDBNO"+"\t")
    for i in range(3,23):
        f.append(header[i]+"\t")
    f.append("X\n")    
   
    a = (len(Hash1),len(sequence))
    pred_Profile=np.zeros(a)
    for i  in range(len(sequence)):  
        if i not in rem:
            T=np.copy(list(sequence))
            T=" ".join(T)
            T=T.split(" ")
            T[i]='[MASK]'
            T=" ".join(T)
            l=fill_mask(T)
            number=len(l)
            for k in range(number):
                token=l[k]['token_str']
                token=token.replace("▁","")
                score=l[k]['score']
                if token not in Hash1:
                    print(i,token)
                
                else:   
                    pred_Profile[Hash1[token]][i]=int(score*100)
            f.append(str(i+1))
            for k in range(len(Hash1)):  #without X
                f.append("\t"+str(pred_Profile[k][i]))
            f.append("\n")
            print(i)
    if len(rem)!=0:
        pred_Profile=np.delete(pred_Profile,rem,1)
    return(pred_Profile)

def Predict_profile1(sequence, header = header,rem=rem,Hash1 = Hash1):
    f=list()
    f.append("PDBNO"+"\t")
    for i in range(3,23):
        f.append(header[i]+"\t")
    f.append("X\n")    
   
    a = (len(Hash1),len(sequence))
    pred_Profile=np.zeros(a)
    for i  in range(len(sequence)):  
        if i not in rem:
            T=np.copy(list(sequence))
            T=" ".join(T)
            T=T.split(" ")
            T[i]='[MASK]'
            T=" ".join(T)
            l=fill_mask(T)
            number=len(l)
            for k in range(number):
                token=l[k]['token_str']
                token=token.replace("▁","")
                score=l[k]['score']
                if token not in Hash1:
                  pred_Profile['X'][i]=pred_Profile['X'][i]+score
                
                else:   
                  pred_Profile[Hash1[token]][i]=score
            f.append(str(i+1))
            for k in range(len(Hash1)):  #without X
                f.append("\t"+str(pred_Profile[k][i]))
            f.append("\n")
            print(i)
    if len(rem)!=0:
        pred_Profile=np.delete(pred_Profile,rem,1)
    return(pred_Profile)

def print_func(sequence):
  s = Predict_profile1(sequence)
  ss = list(s)
  final = []
  for i in range(len(s)):
    # q= np.concatenate((codes[i],s[i]))
    q = [str(codes[i])] + str(ss[i]).replace('[','').replace(']','').split(" ")
    final.append(q)
  res = "\n".join("  ".join(str(el) for el in row) for row in final)
  return res


title="Protein sequence profile prediction using ProtAlbert transformer"
description="""Please enter the sequence.
* Prediction process can take longer for long sequences.
"""



iface = gr.Interface(fn=print_func,
                     inputs=["text"], 
                     outputs="text",
                     description=description,
                     title=title)
iface.launch()