Protein subcellular localization annotations were obtained from UniProtKB (release 2022_02, subcellular location section) (The UniProt Consortium, 2025), Human Protein Atlas (HPA) (version 22, accessed on 9 June 2022) (Thul et al., 2017), and OpenCell (accessed on 10 June 2022) (Cho et al., 2022). Database-specific confidence criteria were applied: UniProt annotations were required to contain any supporting evidence, HPA annotations classified as “Uncertain” were excluded, and OpenCell annotations assigned to its lowest-confidence category were omitted.

For the analysis of protein multiple localization, localization annotations from HPA and UniProt were grouped into the following generalized categories: Nucleus, Cytoplasm, Endomembrane system, Other (including chromosome, secretory, and extracellular proteins in UniProt).

The construction and validation of the SimChrom classification have been described in detail previously (Gribkova et al., 2026). Briefly, the classification structure was developed through the analysis of published literature, specialized databases, and MS-based chromatome studies. Proteins were assigned to the corresponding categories primarily on the basis of Gene Ontology annotations (annotation dataset generated on 6 March 2025) and predefined mappings between Gene Ontology terms and SimChrom categories. The resulting assignments were subsequently supplemented and, where appropriate, filtered using evidence from specialized databases and the literature, followed by manual curation.

To ensure unambiguous categorization, we constructed a SimChrom-SL (single labeled) classification, where each protein was assigned to exactly one category of the same SimChrom ontology. Assignment followed a predefined priority order: “Molecular function” and “Physico-chemical properties” categories were prioritized, followed by others (“Biological processes” and “Genomic location”). Within these groups, categories containing fewer proteins were ordered first. Each protein was labeled with its first eligible category in this sequence.

Published MS-based datasets of human nuclear and chromatin-associated proteins were derived through PubMed search. The selected studies represented total chromatin (Torrente et al., 2011; Kustatscher et al., 2014, Ginno et al., 2018, Shi et al., 2021, Ugur et al., 2023), a core chromatome defined by the recurrent detection of proteins in chromatin fractions across diverse cell and tissue types (Kourtis et al., 2026); nascent chromatin (Alabert et al., 2014; Alvarez et al., 2023), and the nucleome (Itzhak et al., 2016). Hereafter, these datasets are referred to by the first author's name and publication year for ease of identification; the corresponding publications are linked directly in the text.

Protein identifiers reported in the MS-based studies were standardized against the UniProt human reference proteome (release 2022_02). Isoform-specific records were mapped to their corresponding canonical entries, and obsolete UniProt accessions were replaced with identifiers valid in the selected proteome release. Gene symbols were used as secondary identifiers when accession-based mapping was unsuccessful.

From Kustatscher et al., 2014 study, which provides interphase chromatin probability (ICP) scores for 7635 human proteins, proteins with ICP > 0.5 were selected. From Alabert et al., 2014 study entries with missing UniProt ID, gene name, or intensity were excluded, unmapped entries were matched by gene name. No filtering by nascent enrichment or chromatin probability was applied. From Ginno et al., 2018 study proteins quantified with at least two unique peptides and consistent signal across all three replicates of at least one cell cycle stage (G1, S, or M) were selected. Although mitotic chromatin was included, the protein composition across stages was nearly identical, and excluding M-phase would not significantly alter the dataset. From Shi et al., 2021 study, protein entries were taken from the experimental run that used conditions involving HaeIII digestion and no 1,6-hexanediol treatment (“condition 2” in the study), representing chromatin-associated proteins extracted under native conditions using the Hi-MS protocol. Only proteins with nonzero iBAQ values in all three replicates were retained. From Torrente et al., 2011 study chromatin-associated protein lists obtained by three extraction methods were combined (total chromatin extraction, high-salt extraction, and micrococcal nuclease digestion). Protein GI accession numbers were converted to UniProt reviewed entries. From Itzhak et al., 2016 study all entries annotated as “mostly nuclear” by the “Global classifier 2” were taken. From Alvarez et al., 2023 study a combined list of proteins obtained by NCC (Nascent Chromatin Capture) in HeLa S3 cells and iPOND (isolation of Proteins On Nascent DNA) in TIG-3 fibroblasts was taken. From the Ugur et al., 2023 study proteins with nonmissing raw log2 intensity values in all three chromatin replicates for human embryonic stem cells were taken.

Abundance data in the protein table corresponds to the “Whole organism, SC (Peptideatlas,aug,2014)” dataset from PaxDb version 4.2 (Wang et al., 2015).

Protein complex information was downloaded from Complex Portal (Balu et al., 2025) (accessed on 7 January 2025).

Protein domain annotation was obtained from Pfam version 37.0 (Paysan-Lafosse et al., 2025). Only Pfam-A entries classified as Domain, Family, Repeat, or Coiled-coil were included. TED domain annotation was downloaded using TED API v1 (Lau et al., 2024).

Domains engaged in multivalent interactions (EMVI domains) were defined as Pfam domains occurring either in multiple copies or together with another Pfam domain in at least three chromatin regulators. Chromatin regulators comprised nine SimChrom-SL categories: histone chaperones; histone PTM writers, erasers, and readers; histone modification; chromatin remodelers; methylated DNA-binding proteins; DNA (de)methylation; and RNA modification. Based on published evidence, EMVI domains were manually assigned to functional subgroups describing their roles in histone modification, histone or nucleic acid binding, protein interactions, oligomerization, chromatin remodeling, or other processes. Histone modification-related domains were additionally grouped by modification type and classified as writers, erasers, or readers.

Domain co-occurrence was quantified using the conditional probability of detecting domain A in a chromatin protein given the presence of domain B: P(A|B)=P(A ∩ B)/P(B).

To identify potentially novel structural domains and assess the representation of TED domains from chromatin-associated proteins in existing structural databases, we performed structural similarity searches against PDB and CATH50 using the Foldseek easy-search module (van Kempen et al., 2023). The searches were conducted using the exhaustive-search option and otherwise default parameters, with database versions “PDB100 20240101” and “CATH50 4.3.0”. The resulting matches were filtered by a Foldseek probability score greater than 0.9, representing the estimated probability that the query and target are homologous (e.g., belong to the same SCOPe superfamily), and query coverage greater than 0.5. Additional sequence-based searches against CATH were performed using CATH/Gene3D annotations.

TED domains without significant structural matches in either PDB or CATH50 were classified as potentially novel structural domains, which may represent previously uncharacterized structural superfamilies or protein folds. To further characterize these domains, available functional and other annotations were retrieved from InterPro. Only annotations overlapping more than 50% of the corresponding TED domain were retained.