OpenViSTAI · Agents · Blockchain · Crypto
Submit a project
Menu
AI Models

Procyon

ProCyon is an open-source multimodal foundation model for predicting protein phenotypes across biological scales, distributed with a Python implementation, associated model weights and a dataset.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Procyon — OpenViST project profile in AI Models.
OpenViST project illustration. See the linked sources for product details.
Understand the project

Practical overview

When it fits

Use ProCyon when you need a locally configured research model and code examples for protein phenotype generation, retrieval or related protein analyses.

Who it is for

  • Researchers exploring protein phenotype prediction and generation.
  • Teams investigating protein retrieval, binding-related predictions or model evaluation.
  • Developers building local Python research workflows around the provided model and data.

Documented uses

  • Generate protein phenotype predictions with the provided model notebooks.
  • Retrieve proteins or phenotypes for research exploration.
  • Explore drug-binding domain and protein–peptide binding predictions.
  • Evaluate models against baselines using the supplied evaluation framework.
  • Review training examples and scripts for developing model workflows.
Before you choose

Requirements & limits

The repository is public and MIT-licensed. Llama 3 permissions are required for ProCyon-Full and ProCyon-Bind.

  • The documented setup requires CUDA toolkit components, including nvcc.
  • The package depends on ProCyon-Instruct data and associated pretrained weights.
  • For inference, the README recommends transformers 4.31.0 because versions above 4.31.0 change generated outputs.
  • The repository provides distinct Full, Split and Bind model releases, alongside separate notebooks for supported tasks.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Install the ProCyon Python package

Set up the package and its local data and model paths, following the repository's installation instructions.

Before you start

  • A Linux environment with the CUDA toolkit, particularly nvcc.
  • The ProCyon repository and Python; the README recommends uv and also documents pip installation.
  • Access to the ProCyon-Instruct dataset and required pretrained weights, including relevant Llama and ESM2 models.
  • Hugging Face permission for Llama 3 when using ProCyon-Full or ProCyon-Bind.

Steps

  1. From the repository directory, install with uv using the documented build and compile extras, install the package in editable mode, and activate the resulting environment; the README also gives a pip-only installation option.

  2. Obtain the ProCyon-Instruct dataset and associated model weights from their linked Hugging Face locations.

  3. Create a .env file that sets DATA_DIR to the dataset location and HOME_DIR to the repository location; set LLAMA3_PATH as shown when using the Llama 3 model.

Expected outcome

The package is installed and configured with paths to the dataset and model dependencies; the repository links to task notebooks for subsequent use.

Guide limitations

  • This covers setup only; consult the task-specific notebooks for inference or evaluation steps.
  • The repository warns that transformers versions above 4.31.0 change generated outputs and recommends version 4.31.0 for inference.
  • Llama 3 access must be approved for ProCyon-Full and ProCyon-Bind.
Documentation 1 ↗
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
17 October 2026
Original website title
mims-harvard/ProCyon · Official homepage title · retained in its source language
How we review sourcesSuggest a correction