PSALM v2.2.12

Since last release, significant changes to:

1. Data curation and augmentation
2. PSALM model
3. Training
4. Decoding converts residue-level predictions to actual domain calls with start/stop coordinates and confidence scores
5. Evaluation

These changes are reflected in the revised preprint

v1.1.3

v1.1.3 includes better handling and fixes to padding for annotating a FASTA file with multiple sequences. Note that this is the LAST release corresponding to the version of the PSALM preprint dated October 17, 2024.

v1.1.0

v1.1.0 includes more efficient sequence parsing and annotation, additional ways to save predictions at the clan and family level, and bug fixes for visualizations

NOTE: We encourage you to use the PSALM-1b-clan and PSALM-1b-family models, as they are trained on the entirety of Pfam-A Seed 35.0 and can predict across all 19,632 Pfam families and all 655 Pfam clans in Pfam 35.0.

PSALM

This repository contains code and pre-trained weights for Protein Sequence Annotation with Language Models (PSALM) from our 2024 preprint.

Abstract

Protein function inference relies on annotating protein domains via sequence similarity, often modeled through profile Hidden Markov Models (profile HMMs), which capture evolutionary diversity within related domains. However, profile HMMs make strong simplifying independence assumptions when modeling residues in a sequence. Here, we introduce PSALM (Protein Sequence Annotation using Language Models), a hierarchical approach that relaxes these assumptions and uses representations of protein sequences learned by protein language models to enable high-sensitivity, high-specificity residue-level protein sequence annotation. We also develop the Multi-Domain Protein Homology Benchmark (MDPH-Bench), a benchmark for protein sequence domain annotation, where training and test sequences have been rigorously split to share no similarity between any of their domains at a given threshold of sequence identity. Prior benchmarks, which split one domain family at a time, do not support methods for annotating multi-domain proteins, where training and test sequences need to have multiple domains from different families. We validate PSALM’s performance on MDPH-Bench and highlight PSALM as a promising alternative to HMMER, a state-of-the-art profile HMM-based method, for protein sequence annotation.

Usage

PSALM requires Python>=3.10 and PyTorch>=2.2.0. Start a fresh conda environment to use PSALM:

conda create -n "psalm" python=3.10
conda activate psalm
pip install torch protein-sequence-annotation notebook ipykernel
python -m ipykernel install --user

OR just install PSALM alone by using the protein-sequence-annotation PyPI package.

pip install protein-sequence-annotation

After the pip install, you can load and use a pretrained model as follows:

import torch
from psalm import psalm

# Load PSALM clan and fam models
PSALM = psalm(clan_model_name="ProteinSequenceAnnotation/PSALM-1b-clan",
             fam_model_name="ProteinSequenceAnnotation/PSALM-1b-family",
             device = 'cpu') #cpu by default, replace with 'cuda' or 'mps' as needed

# Prepare data (use PSALM.read_fasta(fasta_file_path) to get data directly from a FASTA file)
data = [
    ("Human Beta Globin", "MVHLTPEEKSAVTALWGKVNVDEVGGEALGRLLVVYPWTQRFFESFGDLSTPDAVMGNPKVKAHGKKVLGAFSDGLAHLDNLKGTFATLSELHCDKLHVDPENFRLLGNVLVCVLAHHFGKEFTPPVQAAYQKVVAGVANALAHKYH"),
    ("Flavohemoprotein", "MLDAQTIATVKATIPLLVETGPKLTAHFYDRMFTHNPELKEIFNMSNQRNGDQREALFNAIAAYASNIENLPALLPAVEKIAQKHTSFQIKPEQYNIVGEHLLATLDEMFSPGQEVLDAWGKAYGVLANVFINREAEIYNENASKAGGWEGTRDFRIVAKTPRSALITSFELEPVDGGAVAEYRPGQYLGVWLKPEGFPHQEIRQYSLTRKPDGKGYRIAVKREEGGQVSNWLHNHANVGDVVKLVAPAGDFFMAVADDTPVTLISAGVGQTPMLAMLDTLAKAGHTAQVNWFHAAENGDVHAFADEVKELGQSLPRFTAHTWYRQPSEADRAKGQFDSEGLMDLSKLEGAFSDPTMQFYLCGPVGFMQFTAKQLVDLGVKQENIHYECFGPHKVL")
]

# Visualize PSALM annotations (add optional save_path argument: PSALM.annotate(data,save_path="save_folder")
PSALM.annotate(data)

# Generate predictions without visualization
predictions = PSALM.predict(data)

# Access predictions
for seq_name, pred in predictions.items():
    print(f"Sequence: {seq_name}")
    print("Clan Labels:", pred['clan']['labels'])
    print("Clan Probabilities:", pred['clan']['probs'])
    print("Family Labels:", pred['family']['labels'])
    print("Family Probabilities:", pred['family']['probs'])

Cite

If you find PSALM useful in your research, please cite the following paper:

@article {sarkarkrishnan2024psalm,
	author = {Sarkar, Arpan and Krishnan, Kumaresh and Eddy, Sean R},
	title = {Protein Sequence Domain Annotation using Language Models},
	year = {2024},
	URL = {https://www.biorxiv.org/content/10.1101/2024.06.04.596712v2},
	journal = {bioRxiv}
}

protein-sequence-annotation

PSALM

This package contains code and pre-trained weights for Protein Sequence Annotation with Language Models (PSALM) from our 2024 preprint (June 5th version).

NOTE: PSALM-1 has not been trained on all Pfam families, as it has been trained and benchmarked on highly-curated datasets with strict sequence similarity guarantees between train and test data. PSALM-1b (trained on all families in Pfam 35.0) is coming soon

Abstract

Protein function inference relies on annotating protein domains via sequence similarity, often modeled through profile Hidden Markov Models (profile HMMs), which capture evolutionary diversity within related domains. However, profile HMMs make strong simplifying independence assumptions when modeling residues in a sequence. Here, we introduce PSALM (Protein Sequence Annotation with Language Models), a hierarchical approach that relaxes these assumptions and uses representations of protein sequences learned by protein language models to enable high-sensitivity, high-specificity residue-level protein sequence annotation. We validate PSALM's performance on a curated set of "ground truth" annotations determined by a profile HMM-based method and highlight PSALM as a promising alternative for protein sequence annotation.

Usage

PSALM requires Python>=3.10 and PyTorch>=2.2.0. Start a fresh conda environment to use PSALM:

conda create -n "psalm" python=3.10
conda activate psalm
pip install torch protein-sequence-annotation notebook ipykernel
python -m ipykernel install --user

OR just install PSALM alone by using the protein-sequence-annotation PyPI package.

pip install protein-sequence-annotation

After the pip install, you can load and use a pretrained model as follows:

import torch
from psalm import psalm

# Load PSALM clan and fam models
PSALM = psalm(clan_model_name="ProteinSequenceAnnotation/PSALM-1-clan",
             fam_model_name="ProteinSequenceAnnotation/PSALM-1-family",
             device = 'cpu') #cpu by default, replace with 'cuda' or 'mps' as needed

# Prepare data (use PSALM.read_fasta(fasta_file_path) to get data directly from a FASTA file)
data = [
    ("Human Beta Globin", "MVHLTPEEKSAVTALWGKVNVDEVGGEALGRLLVVYPWTQRFFESFGDLSTPDAVMGNPKVKAHGKKVLGAFSDGLAHLDNLKGTFATLSELHCDKLHVDPENFRLLGNVLVCVLAHHFGKEFTPPVQAAYQKVVAGVANALAHKYH"),
    ("Flavohemoprotein", "MLDAQTIATVKATIPLLVETGPKLTAHFYDRMFTHNPELKEIFNMSNQRNGDQREALFNAIAAYASNIENLPALLPAVEKIAQKHTSFQIKPEQYNIVGEHLLATLDEMFSPGQEVLDAWGKAYGVLANVFINREAEIYNENASKAGGWEGTRDFRIVAKTPRSALITSFELEPVDGGAVAEYRPGQYLGVWLKPEGFPHQEIRQYSLTRKPDGKGYRIAVKREEGGQVSNWLHNHANVGDVVKLVAPAGDFFMAVADDTPVTLISAGVGQTPMLAMLDTLAKAGHTAQVNWFHAAENGDVHAFADEVKELGQSLPRFTAHTWYRQPSEADRAKGQFDSEGLMDLSKLEGAFSDPTMQFYLCGPVGFMQFTAKQLVDLGVKQENIHYECFGPHKVL")
]

# Visualize PSALM annotations (add optional save_path argument: PSALM.annotate(data,save_path="save_folder")
PSALM.annotate(data)

Cite

If you find PSALM useful in your research, please cite the following paper:

@article {sarkarkrishnan2024psalm,
	author = {Sarkar, Arpan and Krishnan, Kumaresh and Eddy, Sean R},
	title = {Protein Sequence Domain Annotation using Language Models},
	year = {2024},
	URL = {https://www.biorxiv.org/content/early/2024/06/05/2024.06.04.596712},
	journal = {bioRxiv}
}