app.py

import os, time, sys


if not os.path.isfile("RF2_apr23.pt"):
    # send param download into background
    os.system(
        "(apt-get install aria2; aria2c -q -x 16 https://colabfold.steineggerlab.workers.dev/RF2_apr23.pt) &"
    )

if not os.path.isdir("RoseTTAFold2"):
    print("install RoseTTAFold2")
    os.system("git clone https://github.com/sokrypton/RoseTTAFold2.git")
    print(os.listdir("RoseTTAFold2"))
    os.system(
        "cd RoseTTAFold2/SE3Transformer; pip -q install --no-cache-dir -r requirements.txt; pip -q install ."
    )
    os.system(
        "wget https://raw.githubusercontent.com/sokrypton/ColabFold/beta/colabfold/mmseqs/api.py"
    )

    # install hhsuite
    print("install hhsuite")
    os.makedirs("hhsuite", exist_ok=True)
    os.system(
        f"curl -fsSL https://github.com/soedinglab/hh-suite/releases/download/v3.3.0/hhsuite-3.3.0-SSE2-Linux.tar.gz | tar xz -C hhsuite/"
    )
    print(os.listdir("hhsuite"))


if os.path.isfile(f"RF2_apr23.pt.aria2"):
    print("downloading RoseTTAFold2 params")
    while os.path.isfile(f"RF2_apr23.pt.aria2"):
        time.sleep(5)

os.environ["DGLBACKEND"] = "pytorch"
sys.path.append("RoseTTAFold2/network")
if "hhsuite" not in os.environ["PATH"]:
    os.environ["PATH"] += ":hhsuite/bin:hhsuite/scripts"

import matplotlib.pyplot as plt
import numpy as np
from parsers import parse_a3m
from api import run_mmseqs2
import torch
from string import ascii_uppercase, ascii_lowercase
import hashlib, re, os
import random

from Bio.PDB import *


def get_hash(x):
    return hashlib.sha1(x.encode()).hexdigest()


alphabet_list = list(ascii_uppercase + ascii_lowercase)
from collections import OrderedDict, Counter

import gradio as gr

if not "pred" in dir():
    from predict import Predictor

    print("compile RoseTTAFold2")
    model_params = "RF2_apr23.pt"
    if torch.cuda.is_available():
        pred = Predictor(model_params, torch.device("cuda:0"))
    else:
        print("WARNING: using CPU")
        pred = Predictor(model_params, torch.device("cpu"))


def get_unique_sequences(seq_list):
    unique_seqs = list(OrderedDict.fromkeys(seq_list))
    return unique_seqs


def get_msa(seq, jobname, cov=50, id=90, max_msa=2048, mode="unpaired_paired"):
    assert mode in ["unpaired", "paired", "unpaired_paired"]
    seqs = [seq] if isinstance(seq, str) else seq

    # collapse homooligomeric sequences
    counts = Counter(seqs)
    u_seqs = list(counts.keys())
    u_nums = list(counts.values())

    # expand homooligomeric sequences
    first_seq = "/".join(sum([[x] * n for x, n in zip(u_seqs, u_nums)], []))
    msa = [first_seq]

    path = os.path.join(jobname, "msa")
    os.makedirs(path, exist_ok=True)
    if mode in ["paired", "unpaired_paired"] and len(u_seqs) > 1:
        print("getting paired MSA")
        out_paired = run_mmseqs2(u_seqs, f"{path}/", use_pairing=True)
        headers, sequences = [], []
        for a3m_lines in out_paired:
            n = -1
            for line in a3m_lines.split("\n"):
                if len(line) > 0:
                    if line.startswith(">"):
                        n += 1
                        if len(headers) < (n + 1):
                            headers.append([])
                            sequences.append([])
                        headers[n].append(line)
                    else:
                        sequences[n].append(line)
        # filter MSA
        with open(f"{path}/paired_in.a3m", "w") as handle:
            for n, sequence in enumerate(sequences):
                handle.write(f">n{n}\n{''.join(sequence)}\n")
        os.system(
            f"hhfilter -i {path}/paired_in.a3m -id {id} -cov {cov} -o {path}/paired_out.a3m"
        )
        with open(f"{path}/paired_out.a3m", "r") as handle:
            for line in handle:
                if line.startswith(">"):
                    n = int(line[2:])
                    xs = sequences[n]
                    # expand homooligomeric sequences
                    xs = ["/".join([x] * num) for x, num in zip(xs, u_nums)]
                    msa.append("/".join(xs))

    if len(msa) < max_msa and (
        mode in ["unpaired", "unpaired_paired"] or len(u_seqs) == 1
    ):
        print("getting unpaired MSA")
        out = run_mmseqs2(u_seqs, f"{path}/")
        Ls = [len(seq) for seq in u_seqs]
        sub_idx = []
        sub_msa = []
        sub_msa_num = 0
        for n, a3m_lines in enumerate(out):
            sub_msa.append([])
            with open(f"{path}/in_{n}.a3m", "w") as handle:
                handle.write(a3m_lines)
            # filter
            os.system(
                f"hhfilter -i {path}/in_{n}.a3m -id {id} -cov {cov} -o {path}/out_{n}.a3m"
            )
            with open(f"{path}/out_{n}.a3m", "r") as handle:
                for line in handle:
                    if not line.startswith(">"):
                        xs = ["-" * l for l in Ls]
                        xs[n] = line.rstrip()
                        # expand homooligomeric sequences
                        xs = ["/".join([x] * num) for x, num in zip(xs, u_nums)]
                        sub_msa[-1].append("/".join(xs))
                        sub_msa_num += 1
            sub_idx.append(list(range(len(sub_msa[-1]))))

        while len(msa) < max_msa and sub_msa_num > 0:
            for n in range(len(sub_idx)):
                if len(sub_idx[n]) > 0:
                    msa.append(sub_msa[n][sub_idx[n].pop(0)])
                    sub_msa_num -= 1
                if len(msa) == max_msa:
                    break

    with open(f"{jobname}/msa.a3m", "w") as handle:
        for n, sequence in enumerate(msa):
            handle.write(f">n{n}\n{sequence}\n")


from Bio.PDB.PDBExceptions import PDBConstructionWarning
import warnings
from Bio.PDB import *
import numpy as np


def add_plddt_to_cif(best_plddts, best_plddt, best_seed, jobname):
    pdb_parser = PDBParser()
    warnings.filterwarnings("ignore", category=PDBConstructionWarning)
    structure = pdb_parser.get_structure(
        "pdb", f"{jobname}/rf2_seed{best_seed}_00_pred.pdb"
    )
    io = MMCIFIO()
    io.set_structure(structure)
    io.save(f"{jobname}/rf2_seed{best_seed}_00_pred.cif")
    plddt_cif = f"""#
loop_
_ma_qa_metric.id
_ma_qa_metric.mode
_ma_qa_metric.name
_ma_qa_metric.software_group_id
_ma_qa_metric.type
1 global pLDDT 1 pLDDT 
2 local  pLDDT 1 pLDDT 
#
_ma_qa_metric_global.metric_id    1
_ma_qa_metric_global.metric_value {best_plddt:.3f}
_ma_qa_metric_global.model_id     1
_ma_qa_metric_global.ordinal_id   1
#
loop_
_ma_qa_metric_local.label_asym_id
_ma_qa_metric_local.label_comp_id
_ma_qa_metric_local.label_seq_id
_ma_qa_metric_local.metric_id
_ma_qa_metric_local.metric_value
_ma_qa_metric_local.model_id
_ma_qa_metric_local.ordinal_id"""

    for chain in structure[0]:
        for i, residue in enumerate(chain):
            plddt_cif += f"\n{chain.id} {residue.resname} {residue.id[1]} 2 {best_plddts[i]*100:.2f} 1 {residue.id[1]}"
    plddt_cif += "\n#"
    with open(f"{jobname}/rf2_seed{best_seed}_00_pred.cif", "a") as f:
        f.write(plddt_cif)


def predict(
    sequence,
    jobname,
    sym,
    order,
    msa_concat_mode,
    msa_method,
    pair_mode,
    collapse_identical,
    num_recycles,
    use_mlm,
    use_dropout,
    max_msa,
    random_seed,
    num_models,
    mode="web",
):
    if not os.path.exists("/home/user/app"):  # crude check if on spaces
        if len(sequence) > 600:
            raise gr.Error(
                f"Your sequence is too long ({len(sequence)}). "
                "Please use the full version of RoseTTAfold2 directly from GitHub."
            )
    random_seed = int(random_seed)
    num_models = int(num_models)
    max_msa = int(max_msa)
    num_recycles = int(num_recycles)
    order = int(order)

    max_extra_msa = max_msa * 8
    sequence = re.sub("[^A-Z:]", "", sequence.replace("/", ":").upper())
    sequence = re.sub(":+", ":", sequence)
    sequence = re.sub("^[:]+", "", sequence)
    sequence = re.sub("[:]+$", "", sequence)

    if sym in ["X", "C"]:
        copies = int(order)
    elif sym in ["D"]:
        copies = int(order) * 2
    else:
        copies = {"T": 12, "O": 24, "I": 60}[sym]
        order = ""
    symm = sym + str(order)

    sequences = sequence.replace(":", "/").split("/")
    if collapse_identical:
        u_sequences = get_unique_sequences(sequences)
    else:
        u_sequences = sequences
    sequences = sum([u_sequences] * copies, [])
    lengths = [len(s) for s in sequences]

    # TODO
    subcrop = 1000 if sum(lengths) > 1400 else -1

    sequence = "/".join(sequences)
    jobname = jobname + "_" + symm + "_" + get_hash(sequence)[:5]

    print(f"jobname: {jobname}")
    print(f"lengths: {lengths}")

    os.makedirs(jobname, exist_ok=True)
    if msa_method == "mmseqs2":
        get_msa(u_sequences, jobname, mode=pair_mode, max_msa=max_extra_msa)

    elif msa_method == "single_sequence":
        u_sequence = "/".join(u_sequences)
        with open(f"{jobname}/msa.a3m", "w") as a3m:
            a3m.write(f">{jobname}\n{u_sequence}\n")

    elif msa_method == "custom_a3m":
        print("upload custom a3m")
        # msa_dict = files.upload()
        lines = msa_dict[list(msa_dict.keys())[0]].decode().splitlines()
        a3m_lines = []
        for line in lines:
            line = line.replace("\x00", "")
            if len(line) > 0 and not line.startswith("#"):
                a3m_lines.append(line)

        with open(f"{jobname}/msa.a3m", "w") as a3m:
            a3m.write("\n".join(a3m_lines))

    best_plddt = None
    best_seed = None
    for seed in range(int(random_seed), int(random_seed) + int(num_models)):
        torch.manual_seed(seed)
        random.seed(seed)
        np.random.seed(seed)
        npz = f"{jobname}/rf2_seed{seed}_00.npz"
        pred.predict(
            inputs=[f"{jobname}/msa.a3m"],
            out_prefix=f"{jobname}/rf2_seed{seed}",
            symm=symm,
            ffdb=None,  # TODO (templates),
            n_recycles=num_recycles,
            msa_mask=0.15 if use_mlm else 0.0,
            msa_concat_mode=msa_concat_mode,
            nseqs=max_msa,
            nseqs_full=max_extra_msa,
            subcrop=subcrop,
            is_training=use_dropout,
        )
        plddt = np.load(npz)["lddt"].mean()
        if best_plddt is None or plddt > best_plddt:
            best_plddt = plddt
            best_plddts = np.load(npz)["lddt"]
            best_seed = seed

        if mode == "web":
            # Mol* only displays AlphaFold plDDT if they are in a cif.
            pdb_parser = PDBParser()
            mmcif_parser = MMCIFParser()

            plddt_cif = add_plddt_to_cif(best_plddts, best_plddt, best_seed, jobname)

            return f"{jobname}/rf2_seed{best_seed}_00_pred.cif"
        else:
            # for api just return a pdb file
            return f"{jobname}/rf2_seed{best_seed}_00_pred.pdb"


def predict_api(
    sequence,
    jobname,
    sym,
    order,
    msa_concat_mode,
    msa_method,
    pair_mode,
    collapse_identical,
    num_recycles,
    use_mlm,
    use_dropout,
    max_msa,
    random_seed,
    num_models,
):
    filename = predict(
        sequence,
        jobname,
        sym,
        order,
        msa_concat_mode,
        msa_method,
        pair_mode,
        collapse_identical,
        num_recycles,
        use_mlm,
        use_dropout,
        max_msa,
        random_seed,
        num_models,
        mode="api",
    )
    with open(f"{filename}") as fp:
        return fp.read()


def molecule(input_pdb, public_link):
    print(input_pdb)
    print(public_link + "/file=" + input_pdb)
    link = public_link + "/file=" + input_pdb
    x = (
        """<!DOCTYPE html>
<html lang="en">
  <head>
    <meta charset="utf-8" />
    <meta name="viewport" content="width=device-width, user-scalable=no, minimum-scale=1.0, maximum-scale=1.0">
    <title>PDBe Molstar - Helper functions</title>
    <!-- Molstar CSS & JS -->
    <link rel="stylesheet" type="text/css" href="https://www.ebi.ac.uk/pdbe/pdb-component-library/css/pdbe-molstar-light-3.1.0.css">
    <script type="text/javascript" src="https://www.ebi.ac.uk/pdbe/pdb-component-library/js/pdbe-molstar-plugin-3.1.0.js"></script>
    <style>
      * {
          margin: 0;
          padding: 0;
          box-sizing: border-box;
      }
      .msp-plugin ::-webkit-scrollbar-thumb {
          background-color: #474748 !important;
      }
      .viewerSection {
        margin: 120px 0 0 0px;
      }
      #myViewer{
        float:left;
        width:100%;
        height: 800px;
        position:relative;
      }
      .btn{
      
                font-family: "Open Sans", sans-serif;
                    display: inline-block;
                    outline: none;
                    cursor: pointer;
                    font-weight: 600;
                    border-radius: 3px;
                    padding: 12px 24px;
                    border: 0;
                    margin:0 10px;
                    line-height: 1.15;
                    font-size: 16px;
                    text-decoration: none;
      }
      .btn-orange{
        background: #ff5000;
        color: #fff;
                    
      }
      .btn-gray{
                    color: #3a4149;
                    background: #e7ebee;
       
      }
      .btn:hover{
      transition: all .1s ease;
                        box-shadow: 0 0 0 0 #fff, 0 0 0 3px #ddd;}
    .text-center{
        display: flex;
        align-items: center;
        justify-content: center;
        padding: 20px 0;
        }
    .flex{
    padding: 10px;
        display: flex;
        align-items: center;
        justify-content: center;
        width:fit-content; 
        }
    .flex svg{ 
        margin-right: 10px;
        width:16px;
        height:16px;
        }
        .flex a{
        margin:0 10px;
        }

    </style>
  </head>
  <body>
  <div class="text-center">
    <a class="btn btn-orange flex" href=\""""
        + link
        + """\" target="_blank"> <svg fill="none" stroke="currentColor" stroke-width="1.5" viewBox="0 0 24 24" xmlns="http://www.w3.org/2000/svg" aria-hidden="true">
  <path stroke-linecap="round" stroke-linejoin="round" d="M19.5 13.5L12 21m0 0l-7.5-7.5M12 21V3"></path>
</svg> <span>CIF File</span></a>
    <a class="btn btn-gray flex" href=\""""
        + link.replace(".cif", ".pdb")
        + """\" target="_blank"> <svg fill="none" stroke="currentColor" stroke-width="1.5" viewBox="0 0 24 24" xmlns="http://www.w3.org/2000/svg" aria-hidden="true">
  <path stroke-linecap="round" stroke-linejoin="round" d="M19.5 13.5L12 21m0 0l-7.5-7.5M12 21V3"></path>
</svg>  <span>PDB File</span></a>
    
  </div>
    <div class="viewerSection">
      <!-- Molstar container -->
      <div id="myViewer"></div>
      
    </div>
    <script>
      //Create plugin instance
      var viewerInstance = new PDBeMolstarPlugin();
  
      //Set options (Checkout available options list in the documentation)
      var options = {
        customData: {
          url: \""""
        + link
        + """\",
          format: "cif"
        },
        alphafoldView: true,
        bgColor: {r:255, g:255, b:255},
        //hideCanvasControls: ["selection", "animation", "controlToggle", "controlInfo"]
      }
      
      //Get element from HTML/Template to place the viewer 
      var viewerContainer = document.getElementById("myViewer");
  
      //Call render method to display the 3D view
      viewerInstance.render(viewerContainer, options);
      
    </script>
  </body>
</html>"""
    )

    return f"""<iframe style="width: 100%; height: 1000px" name="result" allow="midi; geolocation; microphone; camera; 
    display-capture; encrypted-media;" sandbox="allow-modals allow-forms 
    allow-scripts allow-same-origin allow-popups 
    allow-top-navigation-by-user-activation allow-downloads" allowfullscreen="" 
    allowpaymentrequest="" frameborder="0" srcdoc='{x}'></iframe>"""


def predict_web(
    sequence,
    jobname,
    sym,
    order,
    msa_concat_mode,
    msa_method,
    pair_mode,
    collapse_identical,
    num_recycles,
    use_mlm,
    use_dropout,
    max_msa,
    random_seed,
    num_models,
):
    if os.path.exists("/home/user/app"):
        public_link = "https://simonduerr-rosettafold2.hf.space/"
    else:
        public_link = "http://localhost:7860"

    filename = predict(
        sequence,
        jobname,
        sym,
        order,
        msa_concat_mode,
        msa_method,
        pair_mode,
        collapse_identical,
        num_recycles,
        use_mlm,
        use_dropout,
        max_msa,
        random_seed,
        num_models,
        mode="web",
    )

    return molecule(filename, public_link)


with gr.Blocks() as rosettafold:
    gr.Markdown("# RoseTTAFold2")
    gr.Markdown(
        """If using please cite: [manuscript](https://www.biorxiv.org/content/10.1101/2023.05.24.542179v1) 
         <br> Heavily based on [RoseTTAFold2 ColabFold notebook](https://colab.research.google.com/github/sokrypton/ColabFold/blob/main/RoseTTAFold2.ipynb)"""
    )
    with gr.Accordion("How to use in PyMol", open=False):
        gr.Markdown(
            """```os.system('wget https://huggingface.co/spaces/simonduerr/rosettafold2/raw/main/rosettafold_pymol.py')
run rosettafold_pymol.py
rosettafold2 sequence, jobname, [sym, order, msa_concat_mode, msa_method, pair_mode, collapse_identical, num_recycles, use_mlm, use_dropout, max_msa, random_seed, num_models]
color_plddt jobname ```
"""
        )
    sequence = gr.Textbox(
        label="sequence",
        value="PIAQIHILEGRSDEQKETLIREVSEAISRSLDAPLTSVRVIITEMAKGHFGIGGELASK",
    )
    jobname = gr.Textbox(label="jobname", value="test")

    with gr.Accordion("Additional settings", open=False):
        sym = gr.Textbox(label="sym", value="X")
        order = gr.Slider(label="order", value=1, step=1, minimum=1, maximum=12)
        msa_concat_mode = gr.Dropdown(
            label="msa_concat_mode",
            value="default",
            choices=["diag", "repeat", "default"],
        )

        msa_method = gr.Dropdown(
            label="msa_method",
            value="single_sequence",
            choices=[
                "mmseqs2",
                "single_sequence",
            ],  # dont allow custom a3m for now , "custom_a3m"
        )
        pair_mode = gr.Dropdown(
            label="pair_mode",
            value="unpaired_paired",
            choices=["unpaired_paired", "paired", "unpaired"],
        )

        num_recycles = gr.Dropdown(
            label="num_recycles", value="6", choices=["0", "1", "3", "6", "12", "24"]
        )

        use_mlm = gr.Checkbox(label="use_mlm", value=False)
        use_dropout = gr.Checkbox(label="use_dropout", value=False)
        collapse_identical = gr.Checkbox(label="collapse_identical", value=False)
        max_msa = gr.Dropdown(
            choices=["16", "32", "64", "128", "256", "512"],
            value="16",
            label="max_msa",
        )
        random_seed = gr.Textbox(label="random_seed", value=0)
        num_models = gr.Dropdown(
            label="num_models", value="1", choices=["1", "2", "4", "8", "16", "32"]
        )

    btn = gr.Button("Run", visible=False)
    btn_web = gr.Button("Run")

    output_plain = gr.HTML()
    output = gr.HTML()

    btn.click(
        fn=predict_api,
        inputs=[
            sequence,
            jobname,
            sym,
            order,
            msa_concat_mode,
            msa_method,
            pair_mode,
            collapse_identical,
            num_recycles,
            use_mlm,
            use_dropout,
            max_msa,
            random_seed,
            num_models,
        ],
        outputs=output_plain,
        api_name="rosettafold2",
    )
    btn_web.click(
        fn=predict_web,
        inputs=[
            sequence,
            jobname,
            sym,
            order,
            msa_concat_mode,
            msa_method,
            pair_mode,
            collapse_identical,
            num_recycles,
            use_mlm,
            use_dropout,
            max_msa,
            random_seed,
            num_models,
        ],
        outputs=output,
    )


rosettafold.launch()