zero_shot/score_mutant.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import sys\n",
    "sys.path.append('..')\n",
    "from transformers import AutoModelForMaskedLM, AutoTokenizer\n",
    "from prosst.structure.quantizer import PdbQuantizer\n",
    "from Bio import SeqIO\n",
    "import torch\n",
    "import pandas as pd\n",
    "from scipy.stats import spearmanr"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "import os\n",
    "os.environ[\"http_proxy\"] = \"http://127.0.0.1:15777\"\n",
    "os.environ[\"https_proxy\"] = \"http://127.0.0.1:15777\""
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Load ProSST from Hugging Face. \n",
    "(You may need to configure the proxy settings if you are in a region that cannot access the hugging face model.)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "deprot = AutoModelForMaskedLM.from_pretrained(\"AI4Protein/ProSST-2048\", trust_remote_code=True)\n",
    "tokenizer = AutoTokenizer.from_pretrained(\"AI4Protein/ProSST-2048\", trust_remote_code=True)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Load strcuture quantizer"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "processor = PdbQuantizer()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Read protein sequence"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "residue_sequence = str(SeqIO.read('example_data/GRB2_HUMAN_Faure_2021.fasta', 'fasta').seq)\n",
    "    "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Quantize the structure"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "structure_sequence = processor(\"example_data/GRB2_HUMAN_Faure_2021.pdb\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Shift the quantized structure sequence, (for 3 special tokens [CLS], [SEP] and [PAD])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "structure_sequence_offset = [i + 3 for i in structure_sequence]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Prepare model input"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "tokenized_res = tokenizer([residue_sequence], return_tensors='pt')\n",
    "input_ids = tokenized_res['input_ids']\n",
    "attention_mask = tokenized_res['attention_mask']\n",
    "structure_input_ids = torch.tensor([1, *structure_sequence_offset, 2], dtype=torch.long).unsqueeze(0)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Inferece "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "with torch.no_grad():\n",
    "    outputs = deprot(\n",
    "        input_ids=input_ids,\n",
    "        attention_mask=attention_mask,\n",
    "        ss_input_ids=structure_input_ids\n",
    "    )\n",
    "logits = torch.log_softmax(outputs.logits[:, 1:-1], dim=-1).squeeze()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Score mutants"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "df = pd.read_csv(\"example_data/GRB2_HUMAN_Faure_2021.csv\")\n",
    "mutants = df['mutant'].tolist()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "vocab = tokenizer.get_vocab()\n",
    "pred_scores = []\n",
    "for mutant in mutants:\n",
    "    mutant_score = 0\n",
    "    for sub_mutant in mutant.split(\":\"):\n",
    "        wt, idx, mt = sub_mutant[0], int(sub_mutant[1:-1]) - 1, sub_mutant[-1]\n",
    "        pred = logits[idx, vocab[mt]] - logits[idx, vocab[wt]]\n",
    "        mutant_score += pred.item()\n",
    "    pred_scores.append(mutant_score)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Compute the spearman correlation"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "SignificanceResult(statistic=0.6997442598613315, pvalue=0.0)"
      ]
     },
     "execution_count": 12,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "spearmanr(pred_scores, df['DMS_score'])"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "pyg",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.10.13"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}