Skip to content
Finalist

Jaymin

I used Claude to train a protein language model that can generate an extremophilic version of any enzyme, maintaining its function, while allowing it to survive extreme temperatures, pH, salinity, etc.

Code Video Cerebral Valley

Core trick

Fine-tunes an ESM2 adapter with a classifier trained on GTDB extremophile secreted proteins to steer masked protein-language-model generation.

Tags

science

Description

I used Claude to train a protein language model that can generate an extremophilic version of any enzyme, maintaining its function, while allowing it to survive extreme temperatures, pH, salinity, etc. This allows engineering and deploying functional enzymes for applications that require extreme resilience. This project was exceptionally complex. Collecting the training data required several datasets, tools, analyses, and filters (GTDB, GenomeSPOT, signalP6, mmseq2). Training the ESM2 adapter and classifiers required meticulous care to avoid biases in the training data and overfitting. And creating the final masked PL generative pipeline was equally complex to retain enzymatic function, avoid biases, and to generalize across any enzyme (ESM, ProteinMPNN, Swiss-prot, foldseq, mmseqs2, M-CSA, etc). Claude made this possible, both with the coding and with setting up the many mini-experiments I needed to weed out biases and structure the problem correctly. A 6-12-month scoping project was compressed into 5 days. The complex workflow can be tracked in my github repo attached and through my labnotebook (https://github.com/jayman1466/Extremophilic-Protein-Translator/blob/main/labnotebook.md)
jayman1466/Extremophilic-Protein-Translator★ 70 forksPythonMITCreated Jul 7, 2026 · last push Aug 19, 2026

README excerpt

# Extremolith

**GUI frontend at https://ept-portal-g2qcpwpcsa-uc.a.run.app/**
(WARNING: currently this just returns precomputed results of a lignin-degrading
laccase, regardless of the sequence submitted. Need to work out a cheap/scalable
GPU solution for live runs by the general public.)

*Extremophilic Protein Translator — secreted proteins from life at the edge.*

Build a labeled dataset of **secreted proteins from extremophilic organisms**
(drawn from GTDB), for fine-tuning a protein language model (e.g. ESM) and/or
training a classifier that learns the *extremophilic trait* of extracellular
proteins.

## Rationale

Proteins exposed to the extracellular environment (secreted proteins) must
function under whatever conditions the organism inhabits. By collecting
secreted proteins from organisms binned by environment (temperature, pH,
salinity), we get a signal for how sequence adapts to extreme conditions.

Two problems this pipeline is explicitly designed around:

1. **Environment labels are noisy.** An organism's isolation source is only a
   proxy — a cell can be found somewhere it is not thriving. We therefore
   combine the GTDB/NCBI metadata proxy with **GenomeSPOT** genomic predictions
   of optimal temperature, pH, and salinity, and keep a confidence label that
   records whether the two agree.

2. **Clade ≠ trait.** A naive dataset lets a model learn to recognize a
   phylogenetic clade and associate it with an environment, rather than
   learning the adaptive tra

Team