Spaces:
Sleeping
Sleeping
File size: 17,376 Bytes
e1a6cd9 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 |
{
"nbformat": 4,
"nbformat_minor": 0,
"metadata": {
"colab": {
"name": "quickdemo.ipynb",
"provenance": [],
"include_colab_link": true
},
"kernelspec": {
"name": "python3",
"display_name": "Python 3"
},
"language_info": {
"name": "python"
}
},
"cells": [
{
"cell_type": "markdown",
"metadata": {
"id": "view-in-github",
"colab_type": "text"
},
"source": [
"<a href=\"https://colab.research.google.com/github/dauparas/ProteinMPNN/blob/main/colab_notebooks/quickdemo.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
]
},
{
"cell_type": "markdown",
"source": [
"#ProteinMPNN\n",
"This notebook is intended as a quick demo, more features to come!"
],
"metadata": {
"id": "AYZebfKn8gef"
}
},
{
"cell_type": "code",
"source": [
"#@title Setup Model\n",
"import json, time, os, sys, glob\n",
"\n",
"if not os.path.isdir(\"ProteinMPNN\"):\n",
" os.system(\"git clone -q https://github.com/dauparas/ProteinMPNN.git\")\n",
"sys.path.append('/content/ProteinMPNN/vanilla_proteinmpnn')\n",
"\n",
"import matplotlib.pyplot as plt\n",
"import shutil\n",
"import warnings\n",
"import numpy as np\n",
"import torch\n",
"from torch import optim\n",
"from torch.utils.data import DataLoader\n",
"from torch.utils.data.dataset import random_split, Subset\n",
"import copy\n",
"import torch.nn as nn\n",
"import torch.nn.functional as F\n",
"import random\n",
"import os.path\n",
"from protein_mpnn_utils import loss_nll, loss_smoothed, gather_edges, gather_nodes, gather_nodes_t, cat_neighbors_nodes, _scores, _S_to_seq, tied_featurize, parse_PDB\n",
"from protein_mpnn_utils import StructureDataset, StructureDatasetPDB, ProteinMPNN\n",
"\n",
"device = torch.device(\"cuda:0\" if (torch.cuda.is_available()) else \"cpu\")\n",
"model_name=\"v_48_020\" # ProteinMPNN model name: v_48_002, v_48_010, v_48_020, v_48_030, v_32_002, v_32_010; v_32_020, v_32_030; v_48_010=version with 48 edges 0.10A noise\n",
"backbone_noise=0.00 # Standard deviation of Gaussian noise to add to backbone atoms\n",
"\n",
"path_to_model_weights='/content/ProteinMPNN/vanilla_proteinmpnn/vanilla_model_weights' \n",
"hidden_dim = 128\n",
"num_layers = 3 \n",
"model_folder_path = path_to_model_weights\n",
"if model_folder_path[-1] != '/':\n",
" model_folder_path = model_folder_path + '/'\n",
"checkpoint_path = model_folder_path + f'{model_name}.pt'\n",
"\n",
"checkpoint = torch.load(checkpoint_path, map_location=device) \n",
"print('Number of edges:', checkpoint['num_edges'])\n",
"noise_level_print = checkpoint['noise_level']\n",
"print(f'Training noise level: {noise_level_print}A')\n",
"model = ProteinMPNN(num_letters=21, node_features=hidden_dim, edge_features=hidden_dim, hidden_dim=hidden_dim, num_encoder_layers=num_layers, num_decoder_layers=num_layers, augment_eps=backbone_noise, k_neighbors=checkpoint['num_edges'])\n",
"model.to(device)\n",
"model.load_state_dict(checkpoint['model_state_dict'])\n",
"model.eval()\n",
"print(\"Model loaded\")"
],
"metadata": {
"id": "iYDU3ftml2k5",
"cellView": "form"
},
"execution_count": null,
"outputs": []
},
{
"cell_type": "code",
"source": [
"import re\n",
"from google.colab import files\n",
"import numpy as np\n",
"\n",
"#########################\n",
"def get_pdb(pdb_code=\"\"):\n",
" if pdb_code is None or pdb_code == \"\":\n",
" upload_dict = files.upload()\n",
" pdb_string = upload_dict[list(upload_dict.keys())[0]]\n",
" with open(\"tmp.pdb\",\"wb\") as out: out.write(pdb_string)\n",
" return \"tmp.pdb\"\n",
" else:\n",
" os.system(f\"wget -qnc https://files.rcsb.org/view/{pdb_code}.pdb\")\n",
" return f\"{pdb_code}.pdb\"\n",
"\n",
"#@markdown ### Input Options\n",
"pdb='6MRR' #@param {type:\"string\"}\n",
"pdb_path = get_pdb(pdb)\n",
"#@markdown - pdb code (leave blank to get an upload prompt)\n",
"\n",
"designed_chain = \"A\" #@param {type:\"string\"}\n",
"fixed_chain = \"\" #@param {type:\"string\"}\n",
"\n",
"if designed_chain == \"\":\n",
" designed_chain_list = []\n",
"else:\n",
" designed_chain_list = re.sub(\"[^A-Za-z]+\",\",\", designed_chain).split(\",\")\n",
"\n",
"if fixed_chain == \"\":\n",
" fixed_chain_list = []\n",
"else:\n",
" fixed_chain_list = re.sub(\"[^A-Za-z]+\",\",\", fixed_chain).split(\",\")\n",
"\n",
"chain_list = list(set(designed_chain_list + fixed_chain_list))\n",
"\n",
"#@markdown - specified which chain(s) to design and which chain(s) to keep fixed. \n",
"#@markdown Use comma:`A,B` to specifiy more than one chain\n",
"\n",
"#chain = \"A\" #@param {type:\"string\"}\n",
"#pdb_path_chains = chain\n",
"##@markdown - Define which chain to redesign\n",
"\n",
"#@markdown ### Design Options\n",
"num_seqs = 1 #@param [\"1\", \"2\", \"4\", \"8\", \"16\", \"32\", \"64\"] {type:\"raw\"}\n",
"num_seq_per_target = num_seqs\n",
"sampling_temp = \"0.1\" #@param [\"0.1\", \"0.15\", \"0.2\", \"0.25\", \"0.3\"]\n",
"#@markdown - Sampling temperature for amino acids, T=0.0 means taking \n",
"#@markdown argmax, T>>1.0 means sample randomly. Suggested values \n",
"#@markdown 0.1, 0.15, 0.2, 0.25, 0.3. Higher values will lead to more diversity.\n",
"\n",
"\n",
"save_score=0 # 0 for False, 1 for True; save score=-log_prob to npy files\n",
"save_probs=0 # 0 for False, 1 for True; save MPNN predicted probabilites per position\n",
"score_only=0 # 0 for False, 1 for True; score input backbone-sequence pairs\n",
"conditional_probs_only=0 # 0 for False, 1 for True; output conditional probabilities p(s_i given the rest of the sequence and backbone)\n",
"conditional_probs_only_backbone=0 # 0 for False, 1 for True; if true output conditional probabilities p(s_i given backbone)\n",
" \n",
"batch_size=1 # Batch size; can set higher for titan, quadro GPUs, reduce this if running out of GPU memory\n",
"max_length=20000 # Max sequence length\n",
" \n",
"out_folder='.' # Path to a folder to output sequences, e.g. /home/out/\n",
"jsonl_path='' # Path to a folder with parsed pdb into jsonl\n",
"omit_AAs='X' # Specify which amino acids should be omitted in the generated sequence, e.g. 'AC' would omit alanine and cystine.\n",
" \n",
"pssm_multi=0.0 # A value between [0.0, 1.0], 0.0 means do not use pssm, 1.0 ignore MPNN predictions\n",
"pssm_threshold=0.0 # A value between -inf + inf to restric per position AAs\n",
"pssm_log_odds_flag=0 # 0 for False, 1 for True\n",
"pssm_bias_flag=0 # 0 for False, 1 for True\n",
"\n",
"\n",
"##############################################################\n",
"\n",
"folder_for_outputs = out_folder\n",
"\n",
"NUM_BATCHES = num_seq_per_target//batch_size\n",
"BATCH_COPIES = batch_size\n",
"temperatures = [float(item) for item in sampling_temp.split()]\n",
"omit_AAs_list = omit_AAs\n",
"alphabet = 'ACDEFGHIKLMNPQRSTVWYX'\n",
"\n",
"omit_AAs_np = np.array([AA in omit_AAs_list for AA in alphabet]).astype(np.float32)\n",
"\n",
"chain_id_dict = None\n",
"fixed_positions_dict = None\n",
"pssm_dict = None\n",
"omit_AA_dict = None\n",
"bias_AA_dict = None\n",
"tied_positions_dict = None\n",
"bias_by_res_dict = None\n",
"bias_AAs_np = np.zeros(len(alphabet))\n",
"\n",
"\n",
"###############################################################\n",
"pdb_dict_list = parse_PDB(pdb_path, input_chain_list=chain_list)\n",
"dataset_valid = StructureDatasetPDB(pdb_dict_list, truncate=None, max_length=max_length)\n",
"\n",
"chain_id_dict = {}\n",
"chain_id_dict[pdb_dict_list[0]['name']]= (designed_chain_list, fixed_chain_list)\n",
"\n",
"print(chain_id_dict)"
],
"metadata": {
"cellView": "form",
"id": "k4o6w2Y23wxO"
},
"execution_count": null,
"outputs": []
},
{
"cell_type": "code",
"source": [
"#@title RUN\n",
"with torch.no_grad():\n",
" print('Generating sequences...')\n",
" for ix, protein in enumerate(dataset_valid):\n",
" score_list = []\n",
" all_probs_list = []\n",
" all_log_probs_list = []\n",
" S_sample_list = []\n",
" batch_clones = [copy.deepcopy(protein) for i in range(BATCH_COPIES)]\n",
" X, S, mask, lengths, chain_M, chain_encoding_all, chain_list_list, visible_list_list, masked_list_list, masked_chain_length_list_list, chain_M_pos, omit_AA_mask, residue_idx, dihedral_mask, tied_pos_list_of_lists_list, pssm_coef, pssm_bias, pssm_log_odds_all, bias_by_res_all, tied_beta = tied_featurize(batch_clones, device, chain_id_dict, fixed_positions_dict, omit_AA_dict, tied_positions_dict, pssm_dict, bias_by_res_dict)\n",
" pssm_log_odds_mask = (pssm_log_odds_all > pssm_threshold).float() #1.0 for true, 0.0 for false\n",
" name_ = batch_clones[0]['name']\n",
"\n",
" randn_1 = torch.randn(chain_M.shape, device=X.device)\n",
" log_probs = model(X, S, mask, chain_M*chain_M_pos, residue_idx, chain_encoding_all, randn_1)\n",
" mask_for_loss = mask*chain_M*chain_M_pos\n",
" scores = _scores(S, log_probs, mask_for_loss)\n",
" native_score = scores.cpu().data.numpy()\n",
"\n",
" for temp in temperatures:\n",
" for j in range(NUM_BATCHES):\n",
" randn_2 = torch.randn(chain_M.shape, device=X.device)\n",
" if tied_positions_dict == None:\n",
" sample_dict = model.sample(X, randn_2, S, chain_M, chain_encoding_all, residue_idx, mask=mask, temperature=temp, omit_AAs_np=omit_AAs_np, bias_AAs_np=bias_AAs_np, chain_M_pos=chain_M_pos, omit_AA_mask=omit_AA_mask, pssm_coef=pssm_coef, pssm_bias=pssm_bias, pssm_multi=pssm_multi, pssm_log_odds_flag=bool(pssm_log_odds_flag), pssm_log_odds_mask=pssm_log_odds_mask, pssm_bias_flag=bool(pssm_bias_flag), bias_by_res=bias_by_res_all)\n",
" S_sample = sample_dict[\"S\"] \n",
" else:\n",
" sample_dict = model.tied_sample(X, randn_2, S, chain_M, chain_encoding_all, residue_idx, mask=mask, temperature=temp, omit_AAs_np=omit_AAs_np, bias_AAs_np=bias_AAs_np, chain_M_pos=chain_M_pos, omit_AA_mask=omit_AA_mask, pssm_coef=pssm_coef, pssm_bias=pssm_bias, pssm_multi=pssm_multi, pssm_log_odds_flag=bool(pssm_log_odds_flag), pssm_log_odds_mask=pssm_log_odds_mask, pssm_bias_flag=bool(pssm_bias_flag), tied_pos=tied_pos_list_of_lists_list[0], tied_beta=tied_beta, bias_by_res=bias_by_res_all)\n",
" # Compute scores\n",
" S_sample = sample_dict[\"S\"]\n",
" log_probs = model(X, S_sample, mask, chain_M*chain_M_pos, residue_idx, chain_encoding_all, randn_2, use_input_decoding_order=True, decoding_order=sample_dict[\"decoding_order\"])\n",
" mask_for_loss = mask*chain_M*chain_M_pos\n",
" scores = _scores(S_sample, log_probs, mask_for_loss)\n",
" scores = scores.cpu().data.numpy()\n",
" all_probs_list.append(sample_dict[\"probs\"].cpu().data.numpy())\n",
" all_log_probs_list.append(log_probs.cpu().data.numpy())\n",
" S_sample_list.append(S_sample.cpu().data.numpy())\n",
" for b_ix in range(BATCH_COPIES):\n",
" masked_chain_length_list = masked_chain_length_list_list[b_ix]\n",
" masked_list = masked_list_list[b_ix]\n",
" seq_recovery_rate = torch.sum(torch.sum(torch.nn.functional.one_hot(S[b_ix], 21)*torch.nn.functional.one_hot(S_sample[b_ix], 21),axis=-1)*mask_for_loss[b_ix])/torch.sum(mask_for_loss[b_ix])\n",
" seq = _S_to_seq(S_sample[b_ix], chain_M[b_ix])\n",
" score = scores[b_ix]\n",
" score_list.append(score)\n",
" native_seq = _S_to_seq(S[b_ix], chain_M[b_ix])\n",
" if b_ix == 0 and j==0 and temp==temperatures[0]:\n",
" start = 0\n",
" end = 0\n",
" list_of_AAs = []\n",
" for mask_l in masked_chain_length_list:\n",
" end += mask_l\n",
" list_of_AAs.append(native_seq[start:end])\n",
" start = end\n",
" native_seq = \"\".join(list(np.array(list_of_AAs)[np.argsort(masked_list)]))\n",
" l0 = 0\n",
" for mc_length in list(np.array(masked_chain_length_list)[np.argsort(masked_list)])[:-1]:\n",
" l0 += mc_length\n",
" native_seq = native_seq[:l0] + '/' + native_seq[l0:]\n",
" l0 += 1\n",
" sorted_masked_chain_letters = np.argsort(masked_list_list[0])\n",
" print_masked_chains = [masked_list_list[0][i] for i in sorted_masked_chain_letters]\n",
" sorted_visible_chain_letters = np.argsort(visible_list_list[0])\n",
" print_visible_chains = [visible_list_list[0][i] for i in sorted_visible_chain_letters]\n",
" native_score_print = np.format_float_positional(np.float32(native_score.mean()), unique=False, precision=4)\n",
" line = '>{}, score={}, fixed_chains={}, designed_chains={}, model_name={}\\n{}\\n'.format(name_, native_score_print, print_visible_chains, print_masked_chains, model_name, native_seq)\n",
" print(line.rstrip())\n",
" start = 0\n",
" end = 0\n",
" list_of_AAs = []\n",
" for mask_l in masked_chain_length_list:\n",
" end += mask_l\n",
" list_of_AAs.append(seq[start:end])\n",
" start = end\n",
"\n",
" seq = \"\".join(list(np.array(list_of_AAs)[np.argsort(masked_list)]))\n",
" l0 = 0\n",
" for mc_length in list(np.array(masked_chain_length_list)[np.argsort(masked_list)])[:-1]:\n",
" l0 += mc_length\n",
" seq = seq[:l0] + '/' + seq[l0:]\n",
" l0 += 1\n",
" score_print = np.format_float_positional(np.float32(score), unique=False, precision=4)\n",
" seq_rec_print = np.format_float_positional(np.float32(seq_recovery_rate.detach().cpu().numpy()), unique=False, precision=4)\n",
" line = '>T={}, sample={}, score={}, seq_recovery={}\\n{}\\n'.format(temp,b_ix,score_print,seq_rec_print,seq)\n",
" print(line.rstrip())\n",
"\n",
"\n",
"all_probs_concat = np.concatenate(all_probs_list)\n",
"all_log_probs_concat = np.concatenate(all_log_probs_list)\n",
"S_sample_concat = np.concatenate(S_sample_list)"
],
"metadata": {
"id": "xMVlYh8Fv2of",
"cellView": "form"
},
"execution_count": null,
"outputs": []
},
{
"cell_type": "code",
"source": [
"# experimental output\n",
"plt.figure(figsize=(20,5), dpi=100)\n",
"plt.imshow(all_probs_concat.mean(0).T,vmin=0,vmax=1)\n",
"plt.xlabel(\"positions\")\n",
"plt.ylabel(\"amino acids\")\n",
"plt.yticks(range(21),list(alphabet))\n",
"plt.show()"
],
"metadata": {
"id": "4jSKLU3L17Sf"
},
"execution_count": null,
"outputs": []
}
]
} |