Guide to running sequence-structure-function pipeline

Refer to our paper in Bioinformatics for more details.

Setup

• Collection of jupyter notebook scripts demonstrating various aspects of pipeline.
• Conda enviornments required to run pipeline and jupyter notebooks are located in conda_yml.
  • seq_struct_func.yml for steps 1,5-7
  • alphafold2.yml for step 2
  • build environnments with conda env create -f X.yml
  • steps 3 and 4 require pyCHARMM and MMTSB to be installed in seq_struc_func
    • https://academiccharmm.org/documentation/latest/pycharmm
    • https://chemrxiv.org/engage/chemrxiv/article-details/63c6b40230e3e5222e937adc
    • https://github.com/BrooksResearchGroup-UM/pyCHARMM-Workshop
    • http://feig.bch.msu.edu/mmtsb/Main_Page
• Recommended resources: 1 GPU with 10 GB memory and 1-4 CPUs
• Scripts are listed in the order they should be run.

Data: (si_data/) Data necessary for running examples

• asr_seq_annotations.xlsx
  • All enzymes, sequences, and annotations from structure-function pipeline
• extant_msa.fasta
  • Multiple sequence alignment used previously to construct ancestral sequence resurrects
• fasta/
  • Sequences in asr_seq_annotations.xlsx written as fasta format
• pdb_with_fad/
  • Directory containing all AlphaFold2 models with FAD cofactor
• top_dock_pose/
  • Directory cotaining lowest energy poses from minimization in explicit protein
• log_reg_models/
  • Pretrained statsmodels logistic regression models

Toppar: (toppar/) CHARMM Topology and Parameter files

Step 1: (consensus/) Generating consensus sequence hits library

• script/gen_consensus_db.ipynb
  • Create database of consensus sequence hits from AlphaFold2 MSAs

Step 2: (model/) Generating AlphaFold2 Structures

• script/run_alphafold_consensus.ipynb
  • Run example protein with AlphaFold2 using consensus sequence hits

Step 3: (cofactor/) Adding FAD Cofactor

• script/fad.ipynb
  • Add FAD cofactor into generated example protein

Step 4: (dock/) Docking Array of Ligands

• script/fftdock.ipynb
  • Use CHARMM Fast Fourier Transform Docking to get initial positions of ligand
• script/prot_min.ipynb
  • Refine FFT poses in explicit protein representation
• script/cluster.ipynb
  • Cluster poses to select representative poses

Step 5: (pred/) Prediction of Stereochemistry and Reactivity

• script/stereo.ipynb
  • Predict stereochemistry from boltzmann weighted representative poses
• script/reactivity.ipynb
  • Predict reactivity from pose features
• script/vis_pred.ipynb
  • Visuallize predicted poses

Step 6: (msa/) Generate Multiple Sequence Alignment Localized to Binding Site

• script/gen_msa.ipynb
  • Generate Multiple Sequence Alignment
• script/get_bs_ss_residues.ipynb
  • Get set of binding site and second shell residues
• script/slice_msa.ipynb
  • Modify MSA to be limited to binding site and second shell residues

Step 7: (seq_func/) Training Sequence-Function Model and SHAP Analysis

• script/run_automl.ipynb
  • Fit multiple sequence alignment to predicted stereochemistry labels with gradient boosted trees and random forest models
• script/shap_analysis.ipynb
  • Calculate SHAP values for residues and visuallize how residues affect stereochemistry