Chimera

Overview

Chimera is a browser-based bioinformatics suite for sequence analysis, molecular cloning, CRISPR design, personal genomics, and therapeutic protein engineering. All computation runs locally in the browser: no data ever leaves the user’s machine. The repository and release pipeline are the canonical public evidence while the deployment endpoint is being repaired.

It is a client-side workbench with no backend, no persistent state, and no network calls except privacy-respecting Matomo analytics. It is licensed proprietary (Catalyst Group). It is not a medical device and carries prominent disclaimers: results are for educational and research use only, not clinical diagnostics.

What ships (v2.9.0)

28 integrated modules organized into four groups:

Sequence Analysis (SEQ) — 9 modules

GC content with sliding window, nucleotide composition dashboards, automated feature annotation, six-frame ORF detection, full protein translation with hydrophobicity profiling (Kyte-Doolittle), Chou-Fasman secondary structure prediction, protein domain scanning (18-family motif database), RNA secondary structure (Nussinov), and codon usage / CAI with organism-specific optimization (E. coli, yeast, human, CHO).

Molecular Tools (TOOLS) — 11 modules

Restriction digest simulator (25-enzyme database), molecular cloning planner, interactive circular plasmid map, Gibson Assembly designer, CRISPR guide RNA designer (SpCas9, PAM detection, off-target scoring), primer design (nearest-neighbor Tm), pairwise alignment (Needleman-Wunsch + Smith-Waterman), multiple sequence alignment (progressive), dot-matrix comparison, BLAST-like seed-and-extend search, and phylogenetic tree construction (UPGMA / NJ + dN/dS).

Engineering & Genomics (ENG) — 6 modules

Antimicrobial resistance gene scanner (25-entry curated panel: β-lactamases, aminoglycosides, tet/qnr/sul/mcr/van), viral genome analyzer (6-frame ORF, CpG islands, codon bias), microbiome classifier (16S rRNA, Shannon/Simpson diversity), Drug Engineering Lab (4 templates × 8 FDA-approved analogs — Insulin, GLP-1, EPO, IFN-α2b — with click-to-mutate residue editor and heuristic pharmacological property scoring), DNA synthesizer, and mutation simulator.

Personal Genomics (YOU) — 2 modules

Personal genomics suite with 23 clinical report categories and 380+ known variant entries. Auto-detects and parses VCF 4.x, 23andMe raw data, and AncestryDNA exports — all client-side, no upload.

The six-round polish pipeline

Chimera’s v2.9.0 release is the product of six structured polish rounds, each dispatched as a written spec to a specialist model lane:

Round 6 is worth calling out. Fable 5 read every important file in the repo in one pass — all 30+ component files, all 30+ lib modules, the test suite, the CHANGELOG, and the design system — then produced a structured report with file:line evidence. The three critical findings it surfaced (CRISPR score non-determinism from Math.random() in the scoring path, BLAST E-value depending on DB iteration order, and Smith-Waterman affine-gap traceback inconsistency) are real algorithm defects, not style nits. The lib layer was flagged no-touch pending approval, consistent with the standing rule: use Fable for the long-context / coherence lens, Opus for code-heavy reasoning.

Known limitations (from the repo’s own README)

• Client-side only: all algorithms are JavaScript implementations. For production bioinformatics, use established tools (NCBI BLAST, Clustal Omega). Chimera is for rapid exploration and education.
• Simulated pharmacology: the Drug Lab models are heuristic approximations based on published amino acid property scales. Results are directionally informative but not substitutes for molecular dynamics or wet-lab validation.
• Single-threaded: very large sequences (>50 kb) may cause brief UI freezes.
• Not a medical device: not FDA-cleared, CE-marked, or validated for clinical use. All reports are educational.

Why this project matters as an agent project

Chimera is a test bed for the multi-round product polish pipeline that I apply to every serious web surface I build: written spec → specialist model dispatch → atomic commits → test contract → next round. The fact that six rounds shipped clean, with tests passing and a CSS budget enforced, is the pattern. The fact that the sixth round (Fable deep review) found three real algorithm bugs that the previous five rounds missed is the argument for the long-context coherence lens as a distinct step in the pipeline.

The Chagas drug-discovery campaign (2026-07-29) later used Chimera as a read-only audit target — checking whether its modules could be repurposed for T. cruzi research. The audit confirmed Chimera is generic molecular biology software with no organism-specific biology, and that its Drug Lab scoring is a heuristic toy, not a pharmacological model. That honest limitation classification is exactly the kind of evidence I want from my own tools.

Related

• Repo: labs/chimera (GitLab id 83)
• Module reference: docs/MODULE_REFERENCE.md in the repo