Welcome to the GitHub repository for the following publication: Mapping the energetic and allosteric landscapes of protein binding domains (Faure AJ, Domingo J & Schmiedel JM et al., 2022)

Here you'll find an R package with all scripts to reproduce the figures and results from the computational analyses described in the paper.

Table Of Contents

• 1. Required Software
• 2. Installation Instructions
• 3. Required Data
• 4. Pipeline Modes
• 5. Pipeline Stages

Required Software

To run the doubledeepms pipeline you will need the following software and associated packages:

• R >=v3.6.1 (bio3d, Biostrings, coin, Cairo, data.table, ggplot2, GGally, hexbin, plot3D, reshape2, RColorBrewer, ROCR, stringr, ggrepel)

The following software is optional:

• Python v3.8.6 (pandas, numpy, matplotlib, tensorflow, scikit-learn)
• DiMSum v1.2.8 (pipeline for pre-processing deep mutational scanning data i.e. FASTQ to fitness)

Installation Instructions

Open R and enter:

# Install
if(!require(devtools)) install.packages("devtools")
devtools::install_github("lehner-lab/doubledeepms")

# Load
library(doubledeepms)

# Help
?doubledeepms

Required Data

Fitness scores, thermodynamic models, pre-processed data and required miscellaneous files should be downloaded from here and unzipped in your project directory (see '_basedir' option) i.e. where output files should be written.

Pipeline Modes

There are a number of options available for running the doubledeepms pipeline depending on user requirements.

• Basic (default)

Default pipeline functionality ('startStage' = 1) uses prefit thermodynamic models and fitness scores from DMS experiments (already processed with MoCHI and DiMSum respectively; see Required Data) to reproduce all figures in the publication.

• Thermodynamic model inference with MoCHI

Pipeline stage 0 ('doubledeepms_fit_thermo_model') fits thermodynamic models to DMS data for the specified domains ('_tmodelprotein'), using all available data or subsets of phenotypes/variants ('_tmodelsubset'). Parallel computing using job arrays is reccommended while running monte carlo simluations to determine confidence intervals of model-inferred free energies ('_tmodel_jobnumber'). Note: this stage can be resource intensive (up to 48h with 30GB of RAM for GB1).

• Raw read processing

Raw read processing is not handled by the doubledeepms pipeline. FastQ files (GSE184042) from paired-end sequencing of replicate deep mutational scanning (DMS) libraries before ('input') and after selection ('output') were processed using DiMSum (Faure and Schmiedel et al., 2020).

DiMSum command-line arguments and Experimental design files required to obtain variant counts from FastQ files are available here.

Pipeline Stages

The top-level function doubledeepms() is the recommended entry point to the pipeline and by default reproduces the figures and results from the computational analyses described in the following publication: Mapping the energetic and allosteric landscapes of protein binding domains (Faure AJ, Domingo J & Schmiedel JM et al., 2022). See Required Data for instructions on how to obtain all required data and miscellaneous files before running the pipeline.

Stage 0: Fit thermodynamic models with MoCHI

This stage ('doubledeepms_fit_thermo_model') fits thermodynamic models to variant fitness data from (ddPCA) DMS.

Stage 1: Evaluate thermodynamic model results

This stage ('doubledeepms_thermo_model_results') evaluates thermodynamic model results and performance including comparing to literature in vitro measurements (related to Figure 2).

Stage 2: Add 3D structure metrics

This stage ('doubledeepms_structure_metrics') annotates single mutant inferred free energies with PDB structure-derived metrics.

Stage 3: Fitness plots

This stage ('doubledeepms_fitness_plots') plots fitness distributions and scatterplots (related to Figure 1).

Stage 4: Plot fitness heatmaps

This stage ('doubledeepms_fitness_heatmaps') plots single mutant fitness heatmaps (related to Figure 1).

Stage 5: Plot free energy scatterplots

This stage ('doubledeepms_free_energy_scatterplots') plots single mutant free energy scatterplots (related to Figure 3).

Stage 6: Plot free energy heatmaps

This stage ('doubledeepms_free_energy_heatmaps') plots single mutant free energy heatmaps (related to Figure 3).

Stage 7: Protein stability plots

This stage ('doubledeepms_protein_stability_plots') produces protein stability plots (related to Figure 4).

Stage 8: Binding interface plots

This stage ('doubledeepms_interface_mechanisms') produces binding free energy heatmaps for selected GRB2-SH3 residues (related to Figure 5).

Stage 9: Allostery plots

This stage ('doubledeepms_allostery_plots') produces allostery plots (related to Figure 5 and Figure 6).

Stage 10: Allostery scatterplots

This stage ('doubledeepms_allostery_scatterplots') produces free energy scatterplots of major allosteric sites and mutations (related to Figure 5 and Figure 6).

Stage 11: Downsampling analysis

This stage ('doubledeepms_downsampling_analysis') evaluates thermodynamic model results and performance after downsampling (related to Figure 2).

Stage 12: FoldX comparisons

This stage ('doubledeepms_foldx_comparisons') compares inferred folding free energy changes to those predicted by FoldX.

Stage 13: PolyPhen2 comparisons

This stage ('doubledeepms_polyphen2_comparisons') compares inferred folding free energy changes to PolyPhen2 predictions of functional effects.

Stage 14: 3did comparisons

This stage ('doubledeepms_3did_comparisons') tests the enrichment of allosteric mutations at interaction interfaces as annotated by the database of three-dimensional interacting domains (3did).

Stage 15: EVE comparisons

This stage ('doubledeepms_eve_comparisons') compares inferred folding free energy changes to EVE predictions of functional effects.