app.py

import torch
import transformers
from transformers import PreTrainedTokenizerFast
import tranception
import datasets
from tranception import config, model_pytorch
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns
import gradio as gr

tokenizer = PreTrainedTokenizerFast(tokenizer_file="./tranception/utils/tokenizers/Basic_tokenizer",
                                                unk_token="[UNK]",
                                                sep_token="[SEP]",
                                                pad_token="[PAD]",
                                                cls_token="[CLS]",
                                                mask_token="[MASK]"
                                    )
#######################################################################################################################################
###############################################  HELPER FUNCTIONS  ####################################################################
#######################################################################################################################################

AA_vocab = "ACDEFGHIKLMNPQRSTVWY"
def create_all_single_mutants(sequence,AA_vocab=AA_vocab,mutation_range_start=None,mutation_range_end=None):
  all_single_mutants={}
  sequence_list=list(sequence)
  if mutation_range_start is None: mutation_range_start=1
  if mutation_range_end is None: mutation_range_end=len(sequence)
  for position,current_AA in enumerate(sequence[mutation_range_start-1:mutation_range_end]):
    for mutated_AA in AA_vocab:
      if current_AA!=mutated_AA:
        mutated_sequence = sequence_list.copy()
        mutated_sequence[position] = mutated_AA
        all_single_mutants[current_AA+str(position+1)+mutated_AA]="".join(mutated_sequence)
  all_single_mutants = pd.DataFrame.from_dict(all_single_mutants,columns=['mutated_sequence'],orient='index')
  all_single_mutants.reset_index(inplace=True)
  all_single_mutants.columns = ['mutant','mutated_sequence']
  return all_single_mutants

def create_scoring_matrix_visual(scores,sequence,AA_vocab=AA_vocab,mutation_range_start=None,mutation_range_end=None):
  piv=scores.pivot(index='position',columns='target_AA',values='avg_score').transpose().round(4)
  fig, ax = plt.subplots(figsize=(len(sequence)*1.2,20))
  scores_dict = {}
  valid_mutant_set=set(scores.mutant)
  if mutation_range_start is None: mutation_range_start=1
  if mutation_range_end is None: mutation_range_start=len(sequence)
  for target_AA in list(AA_vocab):
    for position in range(mutation_range_start,mutation_range_end+1):
      mutant = sequence[position-1]+str(position)+target_AA
      if mutant in valid_mutant_set:
        scores_dict[mutant]= float(scores.loc[scores.mutant==mutant,'avg_score'])
      else:
        scores_dict[mutant]=0.0
  labels = (np.asarray(["{} \n {:.4f}".format(symb,value) for symb, value in scores_dict.items() ])).reshape(len(AA_vocab),mutation_range_end-mutation_range_start+1)
  heat = sns.heatmap(piv,annot=labels,fmt="",cmap='RdYlGn',linewidths=0.30,vmin=np.percentile(scores.avg_score,2),vmax=np.percentile(scores.avg_score,98),\
              cbar_kws={'label': 'Log likelihood ratio (mutant / starting sequence)'})
  heat.figure.axes[-1].yaxis.label.set_size(20)
  #heat.set_title("Fitness scores for all single amino acid substitutions",fontsize=30)
  heat.set_title("Higher predicted scores (green) imply higher protein fitness",fontsize=30, pad=40)
  heat.set_xlabel("Sequence position", fontsize = 20)
  heat.set_ylabel("Amino Acid mutation", fontsize = 20)
  plt.savefig('fitness_scoring_substitution_matrix.png')
  return plt

def suggest_mutations(scores):
  intro_message = "The following mutations may be sensible options to improve fitness: \n\n"
  #Best mutants
  top_mutants=list(scores.sort_values(by=['avg_score'],ascending=False).head(5).mutant)
  mutant_recos = "The 5 single mutants with highest predicted fitness are:\n {} \n\n".format(", ".join(top_mutants))
  #Best positions
  positive_scores = scores[scores.avg_score > 0]
  positive_scores_position_avg = positive_scores.groupby(['position']).mean()
  top_positions=list(positive_scores_position_avg.sort_values(by=['avg_score'],ascending=False).head(5).index.astype(str))
  print(top_positions)
  position_recos = "The 5 positions with the highest average fitness increase are:\n {}".format(", ".join(top_positions))
  return intro_message+mutant_recos+position_recos

def get_mutated_protein(sequence,mutant):
  mutated_sequence = list(sequence)
  mutated_sequence[int(mutant[1:-1])-1]=mutant[-1]
  return ''.join(mutated_sequence)

def score_and_create_matrix_all_singles(sequence,mutation_range_start=None,mutation_range_end=None,model_type="Small",scoring_mirror=False,batch_size_inference=20,num_workers=0,AA_vocab=AA_vocab):
  if model_type=="Small":
    model = tranception.model_pytorch.TranceptionLMHeadModel.from_pretrained(pretrained_model_name_or_path="PascalNotin/Tranception_Small")
  elif model_type=="Medium":
    model = tranception.model_pytorch.TranceptionLMHeadModel.from_pretrained(pretrained_model_name_or_path="PascalNotin/Tranception_Medium")
  elif model_type=="Large":
    model = tranception.model_pytorch.TranceptionLMHeadModel.from_pretrained(pretrained_model_name_or_path="PascalNotin/Tranception_Large")
  model.config.tokenizer = tokenizer
  all_single_mutants = create_all_single_mutants(sequence,AA_vocab,mutation_range_start,mutation_range_end)
  scores = model.score_mutants(DMS_data=all_single_mutants, 
                                    target_seq=sequence, 
                                    scoring_mirror=scoring_mirror, 
                                    batch_size_inference=batch_size_inference,  
                                    num_workers=num_workers, 
                                    indel_mode=False
                                    )
  scores = pd.merge(scores,all_single_mutants,on="mutated_sequence",how="left")
  scores["position"]=scores["mutant"].map(lambda x: int(x[1:-1]))
  scores["target_AA"] = scores["mutant"].map(lambda x: x[-1])
  score_heatmap = create_scoring_matrix_visual(scores,sequence,AA_vocab,mutation_range_start,mutation_range_end)
  return score_heatmap,suggest_mutations(scores)

#######################################################################################################################################
###############################################  GRADIO INTERFACE  ####################################################################
#######################################################################################################################################

title = "Interactive in silico directed evolution with Tranception"
description = "Perform in silico directed evolution with Tranception to iteratively improve the fitness of a starting protein sequence one mutation at a time. At each step, the Tranception model computes the log likelihood ratios of all possible single amino acid substitution Vs the starting sequence, and outputs a fitness heatmap and recommandations to guide the selection of the mutation to apply. Note: The current version does not currently leverage homologs retrieval at inference time to boost fitness prediction performance."
article = "<p style='text-align: center'><a href='https://proceedings.mlr.press/v162/notin22a.html' target='_blank'>Tranception: Protein Fitness Prediction with Autoregressive Transformers and Inference-time Retrieval</a></p>"
examples=[
['A4_HUMAN: MLPGLALLLLAAWTARALEVPTDGNAGLLAEPQIAMFCGRLNMHMNVQNGKWDSDPSGTKTCIDTKEGILQYCQEVYPELQITNVVEANQPVTIQNWCKRGRKQCKTHPHFVIPYRCLVGEFVSDALLVPDKCKFLHQERMDVCETHLHWHTVAKETCSEKSTNLHDYGMLLPCGIDKFRGVEFVCCPLAEESDNVDSADAEEDDSDVWWGGADTDYADGSEDKVVEVAEEEEVAEVEEEEADDDEDDEDGDEVEEEAEEPYEEATERTTSIATTTTTTTESVEEVVREVCSEQAETGPCRAMISRWYFDVTEGKCAPFFYGGCGGNRNNFDTEEYCMAVCGSAMSQSLLKTTQEPLARDPVKLPTTAASTPDAVDKYLETPGDENEHAHFQKAKERLEAKHRERMSQVMREWEEAERQAKNLPKADKKAVIQHFQEKVESLEQEAANERQQLVETHMARVEAMLNDRRRLALENYITALQAVPPRPRHVFNMLKKYVRAEQKDRQHTLKHFEHVRMVDPKKAAQIRSQVMTHLRVIYERMNQSLSLLYNVPAVAEEIQDEVDELLQKEQNYSDDVLANMISEPRISYGNDALMPSLTETKTTVELLPVNGEFSLDDLQPWHSFGADSVPANTENEVEPVDARPAADRGLTTRPGSGLTNIKTEEISEVKMDAEFRHDSGYEVHHQKLVFFAEDVGSNKGAIIGLMVGGVVIATVIVITLVMLKKKQYTSIHHGVVEVDAAVTPEERHLSKMQQNGYENPTYKFFEQMQN'],
['ADRB2_HUMAN: MGQPGNGSAFLLAPNGSHAPDHDVTQERDEVWVVGMGIVMSLIVLAIVFGNVLVITAIAKFERLQTVTNYFITSLACADLVMGLAVVPFGAAHILMKMWTFGNFWCEFWTSIDVLCVTASIETLCVIAVDRYFAITSPFKYQSLLTKNKARVIILMVWIVSGLTSFLPIQMHWYRATHQEAINCYANETCCDFFTNQAYAIASSIVSFYVPLVIMVFVYSRVFQEAKRQLQKIDKSEGRFHVQNLSQVEQDGRTGHGLRRSSKFCLKEHKALKTLGIIMGTFTLCWLPFFIVNIVHVIQDNLIRKEVYILLNWIGYVNSGFNPLIYCRSPDFRIAFQELLCLRRSSLKAYGNGYSSNGNTGEQSGYHVEQEKENKLLCEDLPGTEDFVGHQGTVPSDNIDSQGRNCSTNDSLL'],
['AMIE_PSEAE: MRHGDISSSNDTVGVAVVNYKMPRLHTAAEVLDNARKIAEMIVGMKQGLPGMDLVVFPEYSLQGIMYDPAEMMETAVAIPGEETEIFSRACRKANVWGVFSLTGERHEEHPRKAPYNTLVLIDNNGEIVQKYRKIIPWCPIEGWYPGGQTYVSEGPKGMKISLIICDDGNYPEIWRDCAMKGAELIVRCQGYMYPAKDQQVMMAKAMAWANNCYVAVANAAGFDGVYSYFGHSAIIGFDGRTLGECGEEEMGIQYAQLSLSQIRDARANDQSQNHLFKILHRGYSGLQASGDGDRGLAECPFEFYRTWVTDAEKARENVERLTRSTTGVAQCPVGRLPYEGLEKEA'],
['P53_HUMAN: MEEPQSDPSVEPPLSQETFSDLWKLLPENNVLSPLPSQAMDDLMLSPDDIEQWFTEDPGPDEAPRMPEAAPRVAPAPAAPTPAAPAPAPSWPLSSSVPSQKTYQGSYGFRLGFLHSGTAKSVTCTYSPALNKMFCQLAKTCPVQLWVDSTPPPGTRVRAMAIYKQSQHMTEVVRRCPHHERCSDSDGLAPPQHLIRVEGNLRVEYLDDRNTFRHSVVVPYEPPEVGSDCTTIHYNYMCNSSCMGGMNRRPILTIITLEDSSGNLLGRNSFEVRVCACPGRDRRTEEENLRKKGEPHHELPPGSTKRALPNNTSSSPQPKKKPLDGEYFTLQIRGRERFEMFRELNEALELKDAQAGKEPGGSRAHSSHLKSKKGQSTSRHKKLMFKTEGPDSD']
]

model_size_selection = gr.Radio(label="Tranception model size (larger models are more accurate but are slower at inference)", choices=["Small","Medium","Large"], value="Small")
protein_sequence_input = gr.Textbox(lines=1, label="Input protein sequence (see below for examples; default = RL40A_YEAST)",value="MQIFVKTLTGKTITLEVESSDTIDNVKSKIQDKEGIPPDQQRLIFAGKQLEDGRTLSDYNIQKESTLHLVLRLRGGIIEPSLKALASKYNCDKSVCRKCYARLPPRATNCRKRKCGHTNQLRPKKKLK")
mutation_range_start = gr.Number(label="Start of mutation range (min value = 1)",value=1,precision=0)
mutation_range_end = gr.Number(label="End of mutation range (leave empty for full lenth)",value=10,precision=0)
scoring_mirror = gr.Checkbox(label="Score protein from both directions (leads to more robust fitness predictions, but doubles inference time)")

#output ==> find a way to make scroallable
output_plot = gr.Plot(label="Fitness scores for all single amino acid substitutions in mutation range")
output_recommendations = gr.Textbox(label="Mutation recommendations")

gr.Interface(
    fn=score_and_create_matrix_all_singles,
    inputs=[protein_sequence_input,mutation_range_start,mutation_range_end,model_size_selection,scoring_mirror],
    outputs=["plot","text"],
    title=title,
    description=description,
    article=article,
    #examples=examples,
    allow_flagging="never"
).launch(debug=True)