# PhyloLoom-AI

- **Event:** [Meta Global AI Developer Hackathon](https://cerebralvalley.ai/e/meta-ai-developer-hackathon)
- **When:** Oct 26, 12:00 PM – Nov 4, 9:00 PM (EST)
- **Where:** Online
- **Team:** [Maliha Khan](https://cerebralvalley.ai/u/mtk_builds)
- **GitHub:** https://github.com/malihakhan1512/PhyloLoom-AI
- **Demo video:** https://youtu.be/NQzu9h-csjI
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4962

PhyloLoom AI is a biological investigation engine that turns protein and DNA sequence observations into transparent, reproducible investigations. It solves a major problem in computational biology tools: results are often presented as black-box scores, AI-generated interpretations, or isolated calculations that users cannot easily verify, reproduce, or trace back to their underlying evidence.

PhyloLoom addresses this by combining a deterministic scientific backend with an evidence-linked AI research copilot. Users can analyze FASTA sequences, compare reference and alternate proteins, detect substitutions and indels, compute physicochemical descriptor changes, explore all 19 alternative amino acids at a selected residue, measure sequence conservation and entropy, identify candidate recurrent substitutions, examine protein-structure contacts, and build exploratory phylogenetic trees.

Every result is classified as COMPUTED, MODEL_PREDICTION, CURATED_ASSERTION, LITERATURE, HYPOTHESIS, or DEMO and connected through an Evidence Graph. Users can click a conclusion and inspect the computation, method, parameters, coordinates, provenance, and checksums behind it. If an analysis was not performed or a model was not verified, PhyloLoom explicitly displays “unavailable” instead of inventing a value or presenting zero as a result.

The platform integrates Meta’s Muse Spark 1.3 model through the Meta Model API. Muse Spark explains computed results and generates falsifiable scientific hypotheses from a structured EvidenceBundle. Every AI-generated statement must cite a valid evidence ID; uncited statements are demoted to suggestions, and hypotheses remain labelled untested until an approved test is performed.

PhyloLoom also addresses common scientific risks such as coordinate confusion between precursor, mature-chain, and submitted-sequence numbering. It never silently converts coordinates and never labels a variant as pathogenic, benign, adaptive, or functionally important without supporting evidence.

The final investigation can be exported as a reproducible JSON or HTML report containing inputs, methods, provenance, limitations, and unavailable analyses.

In short: PhyloLoom AI is not another black-box mutation predictor. It is an inspectable research workspace that helps scientists and students move from sequence observations to evidence-backed, reproducible, and challengeable biological hypotheses—while clearly separating computation, curated knowledge, model predictions, and speculation.

## More from Meta Global AI Developer Hackathon

- [lowbudget](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4959)
- [Jitterbug](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4960)
- [Seal](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4961)
- [424 Devs](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4963)
- [Tolbert Innovation Hub](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4964)
- [h=tvqo5](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4965)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4962. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt