SMSTPred: Protein Language Model Embeddings Predict SAM-Methyltransferase Substrate Specificity and Reveal Metabolite Methyltransferasome Enrichment in Plants and Actinomycetota

S-adenosyl-methionine (SAM)-dependent methyltransferases (MTases) are one of the largest and most ancient enzyme superfamilies in biology, transferring a methyl group from SAM to substrates as different as DNA, RNA, proteins, and small-molecule metabolites. Most SAM-MTases share a Class I Rossmann-like SAM-binding core, with substrate specificity determined by variable inserts and accessory domains rather than the conserved fold; as a result, substrate type is unknown for most MTases in sequence databases, which are dominated by unreviewed entries. Here we present SMSTPred (SAM-Methyltransferase Substrate-Type Predictor), a multi-label classifier that predicts MTase substrate type from protein language model embeddings and benchmarks it under homology-disjoint cross-validation against sequence-homology, structure-homology, and profile-HMM baselines. SMSTPred reaches a cluster-disjoint Matthews correlation coefficient (MCC) of 0.94, significantly ahead of sequence-homology (MMseqs2, 0.85), structure-homology (Foldseek, 0.89) and profile-HMM (0.52) baselines, while providing calibrated, per-substrate confidence. Applying it to ~116,000 MTases across 1,249 reference proteomes, we map the substrate landscape of methylation across the tree of life. Substrate specialization tracks taxonomy: animals expand protein- and DNA-methylation, whereas small-molecule methyltransferases are enriched in plants and Actinomycetota; in Actinomycetota beyond what is expected for their proteome size, and in plants in proportion to their larger proteomes. The predictions identify thousands of uncharacterized methyltransferases with substrate type assignments (about half of those in uncharacterized clusters), many corroborated by an independent reaction-retrieval method. Notably, ~11% of enzymes with a single substrate call lack any class-discriminative sequence motif, suggesting that substrate specificity is often encoded beyond short linear motifs. Together, this work provides SMSTPred, a substrate-type predictor for Class I methyltransferases, and a tree-of-life resource of testable functional hypotheses.

Authors

Institutions

Publication Details

Journal
bioRxiv (Cold Spring Harbor Laboratory)
Published
2026-11-17
DOI
https://doi.org/10.64898/2026.11.13.766537
Primary Topic
Machine Learning in Bioinformatics
Type
preprint

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

SMSTPred: Protein Language Model Embeddings Predict SAM-Methyltransferase Substrate Specificity and Reveal Metabolite Methyltransferasome Enrichment in Plants and Actinomycetota

Adriana Siordia, Dmitrij Rappoport, Adrián Jinich, Ricardo Almada‐Monter et al.
bioRxiv (Cold Spring Harbor Laboratory)
Machine Learning in Bioinformatics
preprint

SMSTPred: Protein Language Model Embeddings Predict SAM-Methyltransferase Substrate Specificity and Reveal Metabolite Methyltransferasome Enrichment in Plants and Actinomycetota

Adriana Siordia, Dmitrij Rappoport, Adrián Jinich, Ricardo Almada‐Monter, Joel Maya Herrera, Nathaniel Roethler
preprint en

Abstract

S-adenosyl-methionine (SAM)-dependent methyltransferases (MTases) are one of the largest and most ancient enzyme superfamilies in biology, transferring a methyl group from SAM to substrates as different as DNA, RNA, proteins, and small-molecule metabolites. Most SAM-MTases share a Class I Rossmann-like SAM-binding core, with substrate specificity determined by variable inserts and accessory domains rather than the conserved fold; as a result, substrate type is unknown for most MTases in sequence databases, which are dominated by unreviewed entries. Here we present SMSTPred (SAM-Methyltransferase Substrate-Type Predictor), a multi-label classifier that predicts MTase substrate type from protein language model embeddings and benchmarks it under homology-disjoint cross-validation against sequence-homology, structure-homology, and profile-HMM baselines. SMSTPred reaches a cluster-disjoint Matthews correlation coefficient (MCC) of 0.94, significantly ahead of sequence-homology (MMseqs2, 0.85), structure-homology (Foldseek, 0.89) and profile-HMM (0.52) baselines, while providing calibrated, per-substrate confidence. Applying it to ~116,000 MTases across 1,249 reference proteomes, we map the substrate landscape of methylation across the tree of life. Substrate specialization tracks taxonomy: animals expand protein- and DNA-methylation, whereas small-molecule methyltransferases are enriched in plants and Actinomycetota; in Actinomycetota beyond what is expected for their proteome size, and in plants in proportion to their larger proteomes. The predictions identify thousands of uncharacterized methyltransferases with substrate type assignments (about half of those in uncharacterized clusters), many corroborated by an independent reaction-retrieval method. Notably, ~11% of enzymes with a single substrate call lack any class-discriminative sequence motif, suggesting that substrate specificity is often encoded beyond short linear motifs. Together, this work provides SMSTPred, a substrate-type predictor for Class I methyltransferases, and a tree-of-life resource of testable functional hypotheses.

bioRxiv (Cold Spring Harbor Laboratory)
University of California, Irvine (US), University of California San Diego (US)
Howard Hughes Medical Institute
Machine Learning in Bioinformatics
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon