Skip to content
Other

kcat-km-esm2-sub4

Enzyme kinetic parameter prediction

2 min readupdated

This bioinformatics model takes an enzyme's amino acid sequence and a substrate's structure to predict the logarithmic values of the turnover number (log10_kcat) and the Michaelis constant (log10_Km). Built by kevinp0, the solution combines two neural network committee branches (a two-tower network built on the ESM2-35M protein language model and RDKit chemical descriptors) with LightGBM gradient boosting and ExtraTrees. Unlike standalone baseline algorithms, this ensemble blends deep protein embeddings, substrate statistics, and sequence composition features while staying within a strict CPU runtime budget. It is aimed at bioinformaticians, computational biologists, and participants in specialized ML competitions.

Key features

  • Hybrid ensemble: The final prediction blends an ESM2-35M neural network committee (0.30 weight), ExtraTrees (0.50), and LightGBM (0.20).
  • Two-tower architecture: Separate branches process protein embeddings (ESM2) and RDKit substrate features (Morgan fingerprints, MACCS keys, and 2D descriptors) before fusing them via [p, l, p*l].
  • Strict CPU runtime control: Inference enforces hard timeouts (finishing within 10 minutes), using an int8-quantized ESM2-35M and a fallback ESM2-8M committee for long sequences.
  • Expanded training dataset: Alongside 42,524 pairs from the primary dataset, training incorporated 27,863 filtered records from the BRENDA and SABIO-RK databases.

What you can do

  • Run batch predictions of log10_kcat and log10_km on enzyme–substrate pairs from a Parquet file and export the results to CSV.
  • Use the pretrained weights in the weights/ folder to reproduce the solution or fine-tune the ensemble for enzyme kinetics tasks.

Example

python predict.py --input INPUT.parquet --output OUTPUT.csv
Read moreDetails
Cross-validation (5 folds, grouped by homology with a 3-mer cosine > 0.25 threshold): Mean SCC = 0.573 (kcat SCC = 0.463, Km SCC = 0.682) vs. 0.383 for the starter ridge baseline.
Hardware and speed: Designed for CPU inference within a 10-minute total budget (up to 150 s for descriptors, up to 430 s for int8 ESM2-35M, and up to 520 s for ESM2-8M).
The repository includes facebook/esm2_t6_8M_UR50D and facebook/esm2_t12_35M_UR50D weights alongside the trained ensemble files.
Discuss Telegram 0

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 0.1 GB · ready files0.7 GB
  • FP8 / INT8recommendedweights 0.1 GB0.6 GB
  • 4-bitweights 0.0 GB0.5 GB

Tip: go 8-bit.FP8 / INT8 is within about a percent of full precision in quality, takes half the memory and runs faster on RTX 40/50 and H100. Full precision only pays off for fine-tuning.

What to get for this model

Massed ComputeOpen Massed Compute

Your own GPU: FP8 / INT8 — from RTX 2060 · BF16 — from RTX 2060

Massed Compute hourly prices, November 2026. The discounted price applies with the coupon at checkout.

Estimate: weights plus working memory. Long videos, high resolutions and long contexts need more.

Run it on Massed Compute

The script downloads the model to the rented machine. How to run it is in the model card.

  1. 1. Rent a machine with a fitting GPU

    Open Massed Compute
  2. 2. Paste the script into the machine's terminal

    Connect over SSH (the Massed Compute panel shows the command) or open a terminal on the machine's desktop.

    HF_TOKEN / CIVITAI_TOKENThe script starts with token lines: paste yours between the quotes. The Hugging Face token is for gated models, the Civitai one for Civitai models.

    # kevinp0/kcat-km-esm2-sub4
    mkdir -p ~/models
    curl -LsSf https://astral.sh/uv/install.sh | sh
    source "$HOME/.local/bin/env"
    uv tool install "huggingface_hub[cli]"
    hf download kevinp0/kcat-km-esm2-sub4 --local-dir ~/models/kcat-km-esm2-sub4
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to kcat-km-esm2-sub4
  3. 3. Use it

    The model is in ~/models. How to run it is in the model card (link above).

Stop the machine in the panel when you are done: it bills by the minute while it runs.

How to speed it up

Picked for this model type. Choose your hardware above to see what fits you exactly.

  • FlashAttention 44.0.0b40from November 18, 2026new release

    fastest exact attention on B200

    Kernels for Blackwell (beta). Exact attention without quantisation — fine for everything, LLMs included.

    How to enable: pip install flash-attn-4; vLLM and SGLang pick it up themselves.

    Earlier generations (2)
    • FlashAttention 3 — The Hopper (H100/H200) version with FP8.
    • FlashAttention 2 — The standard for RTX 30/40/50 and A100: exact attention without the memory cost of long contexts.
  • torch.compile2.15.0from October 28, 2026new release

    10–30% for free

    Compiles the model for your GPU. Slower first run, faster afterwards.

    How to enable: model = torch.compile(model); in ComfyUI the TorchCompileModel node.

Model family

New posts about versions of kcat-km-esm2-sub4 (quants, finetunes, LoRAs) will show up under the bell.

No known relatives of this model yet. All derivatives on Hugging Face

About the model

kevinp0/kcat-km-esm2-sub4
Author
kevinp0
Category
Other
Released
Posted

On NeyroLearn

Hugging Face

00

Ask AI about this model

Answered by Gemini 4 · the chat is kept only in this tab

Comments