nicwulab/HA_Abs

The repository includes all custom scripts and deep learning model code associated with the paper titled "An Explainable Language Model for Antibody Specificity Prediction Using Curated Influenza Hemagglutinin Antibodies.

Sequence analysis of influenza hemagglutinin (HA) antibodies

This README describes the analysis in:
An explainable language model for antibody specificity prediction using curated influenza hemagglutinin antibodies

Contents

• Env setup
• Dataset
• CDR H3 analysis
• mBLM for specificity prediction

Env setup

if you set up env using conda, run conda installation as follow:

conda env create -f Ab_epitope/environment.yml

Dataset

• Flu antibody dataset in this paper: ./doc/HA_Abs_v18.xlsx
• SARS-CoV-2 antibody and HIV dataset is from our previous paper: A large-scale systematic survey reveals recurring molecular features of public antibody responses to SARS-CoV-2
• All antibody from NCBI./doc/all_paired_antibodies_from_GB_v6.xlsx: List of HA antibodies collected from GenBank
• OSA human paired memory B cell sequences: OAS

CDR H3 analysis

1. Extract CDR H3 sequences and references
   python3 script/parse_Ab_table.py

   • Input file:
     • ./doc/HA_Abs_v18.xlsx
   • Output files:
     • ./result/CDRH3.tsv
     • ./result/refs.txt

2. Clustering CDR H3 sequences
   python3 script/CDRH3_clustering_optimal.py

   • Input file:
     • ./result/CDRH3.tsv
   • Output file:
     • ./result/CDRH3_cluster.tsv

3. Analyzing CDR H3 clustering results
   python3 script/analyze_CDRH3_cluster.py

   • Input files:
     • ./doc/HA_Abs_v18.xlsx
     • ./result/CDRH3_cluster.tsv
   • Output files:
     • ./result/Ab_info_CDRH3_clustering.tsv
     • ./result/CDRH3_cluster_summary.tsv

4. Analyzing CDR H3 property
   python3 script/analyze_CDRH3_property.py

   • Input files:
     • ./doc/HA_Abs_v18.xlsx
     • ./doc/all_paired_antibodies_from_GB_v6.xlsx
     • ./result/HA_Abs_clonotype.xlsx
   • Ouput files:
     • ./result/CDRH3_property.tsv
     • ./result/Ab_for_model.tsv

5. Create sequence logos for different CDR H3 clusters
   python3 script/CDRH3_seqlogo.py

   • Input file:
     • ./result/Ab_info_CDRH3_clustering.tsv
   • Output file:
     • ./CDRH3_seqlogo/*.png

6. Plot CDR H3 property for HA head and stem antibodies
   Rscript script/plot_CDRH3_property.R

   • Input file:
     • ./result/CDRH3_property.tsv
   • Output files:
     • ./graph/CDRH3_length.png
     • ./graph/CDRH3_hydrophobicity.png
     • ./graph/tip_hydrophobicity.png

Germline usage analysis

1. Clonotype assignment
   python3 script/assign_clonotype.py

   • Input files:
     • ./doc/HA_Abs_v18.xlsx
     • ./result/CDRH3_cluster.tsv
   • Output file:
     • ./result/HA_Abs_clonotype.xlsx

2. Compute germline usag and extract public clonotype
   python3 script/extract_public_clonotype_VDJ.py

   • Input files:
     • ./doc/HA_Abs_v18.xlsx
     • ./doc/all_paired_antibodies_from_GB_v6.xlsx
     • ./result/HA_Abs_clonotype.xlsx
   • Output files:
     • ./result/IGHV_freq.tsv
     • ./result/IGHD_freq.tsv
     • ./result/IGLV_freq.tsv
     • ./result/IGV_pair_freq.tsv
     • ./result/HA_Abs_clonotype_public.xlsx

3. Extract IGHD4-17-encoded head antibodies
   python3 script/analyze_IGHD4-17.py

   • Input file:
     • ./doc/HA_Abs_v18.xlsx
   • Output file:
     • ./result/Abs_IGHD4-17.tsv

4. Analyzing the occurrence of YGD motif in CDR H3
   python3 script/analyze_YGD_motif.py

   • Input files:
     • ./doc/HA_Abs_v18.xlsx
     • ./doc/all_paired_antibodies_from_GB_v6.xlsx
   • Ouput file:
     • ./result/YGD_motif_freq.tsv

5. Plot VDJ gene usage
   Rscript script/plot_VDJgene_freq.R

   • Input files:
     • ./result/IGHV_freq.tsv
     • ./result/IGHD_freq.tsv
     • ./result/IGLV_freq.tsv
   • Output files:
     • ./graph/IGHV_usage.png
     • ./graph/IGHD_usage.png
     • ./graph/IGLV_usage.png

6. Plot IGHV/IGK(L)V pairing frequency
   Rscript script/plot_Vpair_heatmap.R

   • Input files:
     • ./result/IGV_pair_freq.tsv
   • Output file:
     • ./graph/Vpair_heatmap.png

7. Plot frequency of YGD motif
   Rscript script/plot_YGD_freq.R

   • Input file:
     • ./result/YGD_motif_freq.tsv
   • Output file:
     • ./graph/YGD_motif_freq.png

mBLM for specificity prediction

See Ab_epitope