File: genomopipe.webm (945 KB, 1420x928)

945 KB WEBM

I've set up a bioinformatics toolchain that I run on my arch linux machine.

The workflow is managed by two shell scripts, one does a fetch, and is called by the other that runs the tools.
you input the organism name in the command, then it runs for a long time and outputs reference genomes, annotations, filtered protein sequences, designed backbones + sequences + predicted structures, validation results, logs, and a README summary
The steps it goes through and the tools are:
Genome download (RefSeq preferred to GenBank to clustered fallback) + header cleanup
Quality control (adapter trimming, basic stats) BBTools (bbduk)
Repeat masking RepeatModeler, RepeatMasker
Gene prediction: BRAKER3 for eukaryotes (OrthoDB + optional RNA hints) or Prokka for prokaryotes
Protein extraction/filtering (longest isoform, ≥100 aa prok / >200 aa euk) gffread and biopython
Backbone generation with RFdiffusion (de novo or motif-scaffolded)
Sequence design using ProteinMPNN (8 variants per backbone)
Batch structure prediction with ColabFold (AlphaFold2)
Remote BLASTp validation against nr (top hit recorded)
Functional renaming based on BLAST inference
Markdown report
(managed in Conda envs)
The toolchain automates the entire process from a name through reference genome retrieval, quality control and annotation, protein extraction, AI-driven de novo protein design (backbones and sequences), structure prediction, BLAST-based validation, functional renaming, and generation of a clear Markdown report. Also an electron gui for directing other programs to visualize the output, making queries in a search bar, and a fallback 3dmol.js viewer.