Boltz Pulldown
An experimental AlphaPulldown-like protocol using Boltz for multimer predictions.
Overview
Boltz Pulldown runs multimer predictions from sequence for all target-binder pairs, allowing you to assess binding interactions between multiple potential interaciton partners.
Command-line Options
See available options with --help:
nextflow run boltz_pulldown.nf --help
Multiple Sequence Alignments (MSAs)
By default, no MSAs are used. To enable:
--create_target_msa=true: Generate MSAs for targets--create_binder_msa=true: Generate MSAs for binders
⚠️ For de novo designed binders, use
--create_target_msa=trueto improve target prediction accuracy, but skip generating binder MSAs (--create_binder_msa=false, the default) as these are less informative since no homologs exist.
MSA Generation Options
Option 1: Remote Server
--use_msa_server
Uses the remote ColabFold mmseqs2 server.
Option 2: Local Databases
Download and set up databases with the ColabFold setup_databases.sh script, then specify paths:
--uniref30 /path/to/uniref30
--colabfold_envdb /path/to/colabfold_envdb
Custom Boltz Arguments
Add extra Boltz command-line arguments via ext.args in nextflow.config:
process {
withName: BOLTZ {
accelerator = 1
time = 2.hours
memory = '8g'
cpus = 2
// Example: CPU-only mode
ext.args = "--accelerator cpu"
}
}
Output
Default output directory: results/boltz_pulldown/
Includes:
- Boltz predictions with scores and structures
boltz_pulldown.tsv- summary tableboltz_pulldown_report.html- report with statistics and ipTM score plots