Profluent-Bio/protein2pam-cas12

Protein2PAM models and training data

Overview

This repo contains Protein2PAM models and training data.

For Python API, see:

• GitHub repo: https://github.com/Profluent-AI/protein2pam
• Zenodo archive: https://zenodo.org/records/17843038

Model Summary

PAM prediction models for CRISPR-Cas nucleases.

Main Models

Model Name	Input Protein/Domain	CRISPR Type	Samples
cas8	Cas8 or Cas10d	Type I	28,410
cas9	Cas9 PI-domain	Type II	15,843
cas12	Cas12 protein	Type V	1,720

Additional Models

Model Name	Input Protein/Domain	CRISPR Type	Samples
cas9_full	Cas9 protein	Type II	15,843
cas9_full_nolit	Cas9 protein	Type II	15,731
cas9_pid_nolit	Cas9 PI-domain	Type II	15,731
cas9_pid_nme	Cas9 PI-domain	Type II	15,843
cas12_no_lit	Cas12 protein	Type V	1,675

Dataset Summary

Training data is stored in: Profluent-Bio/protein2pam-training-data

Data Fields

Field name	Description	Data type	Example
cas_family	CRISPR-Cas family classification	string	Cas8
source	Data source ('CRISPR-Cas Atlas' or Literature')	string	CRISPR-Cas Atlas
protein_id	Identifier for literature sequences	string or null	null
citation	Literature citation(s)	string or null	null
doi	DOI for the associated publication	string or null	null
protein_sequence	Full amino acid sequence	string	MTFMILQALYRY...NQN
pid_sequence	PAM-interacting domain sequence (Cas9 only)	string or null	null
pam_consensus	Consensus PAM sequence	string	TTC
pam_logo_acgt	ACGT-ordered numerical matrix for PAM logo	array of float arrays	[[0.005, 0.008, 0.007, 0.01], ...]
type	CRISPR-Cas type classification	string	Type I
subtype	CRISPR-Cas subtype or effector family	string	Cas8

Licensing

Models and data are licensed under the Creative Commons Attribution–NonCommercial 4.0 International (CC BY-NC 4.0). You may share and adapt the models for non-commercial use with appropriate attribution.

Citation

If you use Protein2PAM in your research, please cite the following preprint:
Nayfach, S., Bhatnagar, A., Novichkov, A., et al. (2025). Engineering of CRISPR-Cas PAM recognition using deep learning of vast evolutionary data. bioRxiv.