Atlas Overview

This section introduces the FGMB Atlas resource and summarizes the modeling strategy used to connect molecular QTL resources, prediction models, regulome-wide association studies (RWAS) association testing, and downstream causal fine-mapping.

In this documentation, a molecular trait refers to a source-specific molecular measurement panel used for predictor training. Each molecular trait is defined by both its biological context and molecular modality, such as a brain region, cell type, or cell subtype measured for gene expression, protein abundance, or splicing.

Core resource components include:

• Molecular QTL-derived prediction models for genetically regulated molecular traits.

• RWAS-ready summary resources for disease and aging-related brain traits.

• Cross-context annotations that connect signals across modalities, cohorts, and biological settings.

• Figure and workflow notebooks that document the manuscript analyses.

Molecular Modalities

The current FGMB atlas includes the following molecular modalities:

• Gene expression

• Protein abundance

• Splicing regulation

Within the atlas, each molecular trait is defined as a unique context–modality combination, and these data are used to train genetically regulated expression prediction models for downstream RWAS and causal TWAS analyses.

ROSMAP Resources

ROSMAP provides the largest and most diverse set of molecular reference panels in FGMB. These data include both bulk tissue and single-nucleus resources across multiple brain contexts and modalities.

Bulk tissue RNA-seq

Bulk RNA-seq gene expression data were incorporated from three ROSMAP brain regions:

• Dorsolateral prefrontal cortex (DLPFC), N = 777

• Posterior cingulate cortex (PCC), N = 441

• Anterior cingulate cortex (AC), N = 593

Monocyte gene expression

FGMB also includes a peripheral blood-derived monocyte gene expression panel from ROSMAP:

• Monocyte (Mono), N = 226

ROSMAP protein abundance

Protein abundance data from ROSMAP were incorporated for:

• DLPFC protein abundance (pQTL), N = 416

ROSMAP splicing regulation

Splicing enrichment data derived from bulk RNA-seq were included for three ROSMAP brain regions:

• DLPFC splicing regulation (sQTL), N = 806

• PCC splicing regulation (sQTL), N = 449

• AC splicing regulation (sQTL), N = 603

ROSMAP single-nucleus RNA-seq resources

FGMB includes several ROSMAP-derived single-nucleus RNA-seq resources from dorsolateral prefrontal cortex.

1. CUIMC1 single-nucleus resource

The CUIMC1 dataset includes six major pseudo-bulk cell types of DLPFC region.

• Astrocytes (Ast), N = 419

• Inhibitory neurons (Inh), N = 419

• Excitatory neurons (Exc), N = 419

• Oligodendrocytes (Oli), N = 419

• Oligodendrocyte progenitor cells (OPC), N = 418

• Microglia (Mic), N = 419

2. MIT single-nucleus resource

The MIT dataset includes major cell types and selected subtypes, with sample sizes ranging from 80 to 387:

• Astrocytes (Ast), N = 385

• Inhibitory neurons (Inh), N = 379

• Excitatory neurons (Exc), N = 386

• Oligodendrocytes (Oli), N = 387

• Oligodendrocyte progenitor cells (OPC), N = 383

• Microglia (Mic), N = 377

• Astrocyte subtype 10 (Ast_10), N = 113

• Microglia subtype 12 (Mic_12), N = 106

• Microglia subtype 13 (Mic_13), N = 80

3. Mega single-nucleus resource

The mega-analysis resource integrates multiple single-nucleus datasets and includes six major cell types with sample sizes ranging from 735 to 737:

• Astrocytes (Ast), N = 737

• Inhibitory neurons (Inh), N = 736

• Excitatory neurons (Exc), N = 737

• Oligodendrocytes (Oli), N = 737

• Oligodendrocyte progenitor cells (OPC), N = 735

• Microglia (Mic), N = 733

Mount Sinai Brain Bank (MSBB)

The Mount Sinai Brain Bank contributes bulk tissue molecular reference panels to FGMB. These data broaden regional representation beyond ROSMAP and add complementary bulk brain contexts for expression and protein prediction modeling.

1. MSBB gene expression panels Bulk RNA-seq gene expression data were included from four cortical brain regions:

• Frontal pole (FP, Brodmann area 10), N = 274

• Superior temporal gyrus (STG, Brodmann area 22), N = 254

• Parahippocampal gyrus (PHG, Brodmann area 36), N = 230

• Inferior frontal gyrus (IFG, Brodmann area 44), N = 256

2. MSBB protein abundance panel FGMB also includes protein abundance data from:

• Parahippocampal gyrus (PHG pQTL), N = 184

Knight-ADRC

The Knight Alzheimer’s Disease Research Center contributes parietal cortex molecular reference panels to FGMB. These data add an independent aging-brain cohort and broaden atlas coverage across both cohort and modality.

1. Knight-ADRC gene expression panel Gene expression reference data were included for:

• Parietal cortex (PC eQTL), N = 354

2. Knight-ADRC protein abundance panel Protein abundance reference data were included for:

• Parietal cortex (PC pQTL), N = 412

FGMB Molecular Trait Summary

Table 1 Summary of FGMB molecular trait prediction models.

Source Dataset	Molecular Dataset	Context	Molecular Modality	Sample Size	Genes Trained	Imputable Genes
ROSMAP (De Jager et al. 2018)	DLPFC eQTL	Dorsolateral Prefrontal Cortex	Gene Expression	777	16,307	9,948
	PCC eQTL	Posterior Cingulate Cortex	Gene Expression	441	16,110	10,720
	AC eQTL	Anterior Cingulate Cortex	Gene Expression	593	16,104	10,918
	Mono eQTL	Monocyte	Gene Expression	226	12,801	5,397
ROSMAP (Bennett et al. 2018)	DLPFC pQTL	Dorsolateral Prefrontal Cortex	Protein Expression	416	7,396	3,925
ROSMAP (Najar et al. 2025)	DLPFC sQTL	Dorsolateral Prefrontal Cortex	Splicing Enrichment	806	12,474	7,261
	PCC sQTL	Posterior Cingulate Cortex	Splicing Enrichment	449	12,663	9,832
	AC sQTL	Anterior Cingulate Cortex	Splicing Enrichment	603	12,585	8,472
ROSMAP (Fujita et al. 2024)	Ast eQTL CUIMC1	Astrocyte	Gene Expression	419	11,392	6,023
	Inh eQTL CUIMC1	Inhibitory Neuron	Gene Expression	419	11,266	6,609
	Exc eQTL CUIMC1	Excitatory Neuron	Gene Expression	419	11,111	8,085
	Oli eQTL CUIMC1	Oligodendrocyte	Gene Expression	419	10,912	5,595
	OPC eQTL CUIMC1	Oligodendrocyte Progenitor	Gene Expression	418	7,742	3,582
	Mic eQTL CUIMC1	Microglia	Gene Expression	419	7,130	3,071
ROSMAP (Comandante-Lou et al. 2025)	Ast eQTL MIT	Astrocyte	Gene Expression	385	9,125	5,530
	Ast.10 eQTL	Astrocyte Subtype 10	Gene Expression	113	1,927	1,927
	Inh eQTL MIT	Inhibitory Neuron	Gene Expression	379	10,760	7,141
	Exc eQTL MIT	Excitatory Neuron	Gene Expression	386	10,645	8,000
	Oli eQTL MIT	Oligodendrocyte	Gene Expression	387	10,021	6,748
	OPC eQTL MIT	Oligodendrocyte Progenitor	Gene Expression	383	8,707	5,163
	Mic eQTL MIT	Microglia	Gene Expression	377	5,404	3,306
	Mic.12 eQTL MIT	Microglia Subtype 12	Gene Expression	106	702	701
	Mic.13 eQTL MIT	Microglia Subtype 13	Gene Expression	80	692	692
ROSMAP (Comandante-Lou et al. 2025)	Ast eQTL mega	Astrocyte	Gene Expression	737	7,742	4,428
	Inh eQTL mega	Inhibitory Neuron	Gene Expression	736	9,577	5,970
	Exc eQTL mega	Excitatory Neuron	Gene Expression	737	10,138	7,856
	Oli eQTL mega	Oligodendrocyte	Gene Expression	737	8,196	4,826
	OPC eQTL mega	Oligodendrocyte Progenitor	Gene Expression	735	5,897	2,787
	Mic eQTL mega	Microglia	Gene Expression	733	3,514	1,562
MSBB (Wang et al. 2018)	FP eQTL	Frontal Pole	Gene Expression	274	9,275	8,088
	STG eQTL	Superior Temporal Gyrus	Gene Expression	254	9,275	7,645
	PHG eQTL	Parahippocampal Gyrus	Gene Expression	230	9,275	7,494
	IFG eQTL	Inferior Frontal Gyrus	Gene Expression	256	9,275	7,997
	PHG pQTL	Parahippocampal Gyrus	Protein Expression	184	11,224	8,325
Knight-ADRC (Fernandez et al. 2024)	PC eQTL	Parietal Cortex	Gene Expression	354	15,941	9,070
	PC pQTL	Parietal Cortex	Protein Expression	412	1,018	233

Molecular Dataset refers to a source-specific molecular measurement panel used for predictor training, such as a brain region, cell type, or cell subtype measured for gene expression, protein abundance, or splicing. Context denotes the tissue, cell type, or cell subtype, whereas Molecular Modality denotes the type of molecular phenotype being modeled.