multiomics

A fast, parallel multi-omics analysis CLI written in Rust. Feed it genomics (VCF), transcriptomics (TSV), epigenomics (BED), proteomics (mzML), single-cell RNA (10x MEX), ATAC-seq, CNV, and long-reads — get back an integrated HTML report, a MultiQC-compatible JSON file, and a live terminal dashboard.

┌─ multiomics ─────────────────────────────────────────────────────────────────┐
│  Phase: Genomics Analysis                               Elapsed: 00:01:23    │
├─────────────────────────────────┬────────────────────────────────────────────┤
│  GENOMICS      [████████░░] 82% │  LIVE INSIGHTS                             │
│  58,432 rec/s  ETA: 00:00:12   │  [INFO]  Ti/Tv = 2.14 (normal range)       │
│                                 │  [CRIT]  KRAS high-impact variants         │
│  TRANSCRIPTOMICS [░░░░░░░░]  0% │  [WARN]  TMB-H: 14.2 mut/Mb → pembrolizumab│
│  EPIGENOMICS     [░░░░░░░░]  0% │  [WARN]  MSI-H: mismatch repair deficient  │
│  INTEGRATION     [░░░░░░░░]  0% │  [WARN]  APOBEC mutagenesis (SBS2/SBS13)   │
└──────────────────────────────────────────────────────────────────────────────┘

---

Features

Multi-omics analysis modules

Module	Input	What it computes
Genomics	VCF	Ti/Tv ratio, variant density per chromosome, AF histogram, high-impact variant genes, unique position count (HyperLogLog)
Transcriptomics	TSV (genes × samples)	Expressed gene count, top-100 by TPM, log₂ fold-change DE (when ≥ 2 samples), Welch t-test + BH FDR correction
Epigenomics	BED (bisulfite)	Global methylation %, CpG island detection (Gardiner-Garden 1987 criteria), hyper/hypomethylated regions
Proteomics	mzML + FASTA	Database search, PSM/peptide/protein counts at user-defined FDR, phosphoproteomics
Single-cell RNA	10x MEX directory	QC metrics, log-normalisation, HVG selection, GPU-accelerated UMAP (wgpu k-selection shader)
ATAC-seq	narrowPeak BED6+4	Peak count, median signal, per-chromosome chromatin accessibility
CNV	VCF with CN field	Segment count, fraction genome altered, estimated ploidy
Gene quantification	BAM + GTF	Read-level counting, assignment rate, top expressed genes
FASTQ QC	FASTQ	Read count, GC%, Q30%
Integration	all of the above	PCA, MOFA+ joint factor analysis, cross-modality Pearson correlation, KEGG + GMT pathway enrichment, rule-based insights

Cancer genomics — clinical biomarkers

Analysis	What it does	Clinical relevance
Tumor Mutational Burden (TMB)	total_variants / genome_Mb; auto-detects WGS (2800 Mb) or WES (35 Mb)	FDA-approved pembrolizumab biomarker (TMB-H ≥10 mut/Mb; Chalmers 2017)
Microsatellite Instability (MSI)	Homopolymer indel fraction + short-indel fraction composite score	FDA-approved pembrolizumab biomarker (MSI-H; Bonneville 2017)
COSMIC Mutational Signatures	6-channel SBS spectrum (C>A/G/T, T>A/C/G); detects SBS1/2/4/6/7/13/17/18	Identifies etiology: APOBEC, tobacco, MMR-D, UV, aging (Alexandrov 2020)
Immune Evasion Score	Expression of CD274 (PD-L1), CTLA4, LAG3, HAVCR2 (TIM-3), TIGIT, PDCD1 + B2M antigen presentation	Identifies checkpoint inhibition and antigen loss (Chen & Mellman 2017)
Polygenic Risk Score (PRS)	37 GWAS variants (p<5×10⁻⁸) across 6 cancers; Z-scored vs. population baseline	Germline cancer risk for CRC, breast, lung, prostate, melanoma, ovarian cancer
Tumor Purity	VAF mode × 2 (diploid model) + methylation depletion cross-validation; consensus estimate	Estimates tumor cell fraction (Carter 2012 / ABSOLUTE-inspired)
Kataegis	≥6 consecutive mutations with geometric mean inter-mutation distance < 1000 bp	Hypermutation foci, APOBEC/AID mutagenesis (Alexandrov 2013)
Homologous Recombination Deficiency (HRD)	Indel size spectrum (del 1bp / 2–5bp / 6–50bp / ins >3bp); optional microhomology scoring with --reference	HRD-HIGH indels suggest BRCA1/2 deficiency, PARP inhibitor sensitivity (Watkins 2020)
Loss of Heterozygosity (LOH)	Median |AF − 0.5| per chromosome in heterozygous variants	Detects chromosomal arm LOH (≥10 het variants, median deviation >0.15)

Epigenomics extras

Analysis	What it does	Reference
Horvath Epigenetic Clock	353-CpG DNAm age estimate using anti_trafo transform; confidence: HIGH/MODERATE/LOW	Horvath 2013 (Genome Biology)

Integration extras

Analysis	What it does	Reference
Multi-modal paradox detection	Identifies genes active in one modality but silenced in another	Roadmap Epigenomics 2015
Gene regulatory state classification	Labels each gene as Active/Silenced/Poised/Bivalent/VariantDriven/Paradoxical/Unknown	Roadmap Epigenomics 2015
MOFA+ joint factor analysis	Multi-omics factor analysis across all modalities; extracts latent shared variance	Argelaguet 2018 (Molecular Systems Biology)
GSEA pre-ranked	Gene set enrichment on DE log₂FC ranking	Subramanian 2005

---

Installation

Requirements

• Rust 1.78 or newer — install via rustup.rs
• A C linker (usually already present; on Windows install Build Tools for Visual Studio)

Build (CPU only — works everywhere)

git clone https://github.com/diladeniz/multiomics.git
cd multiomics
cargo build --release --bin multiomics
# binary at: target/release/multiomics

Build with GPU acceleration

GPU support uses wgpu and runs on any modern GPU via Vulkan (Linux/Windows), Metal (macOS), or DX12 (Windows). No CUDA required.

It accelerates the k-nearest-neighbour step of UMAP for single-cell datasets with 5 000+ cells.

cargo build --release --bin multiomics --features gpu

Build everything (ATAC + CNV + longread + GPU)

cargo build --release --bin multiomics --features full,gpu

Install to PATH

cargo install --path cli                    # CPU only
cargo install --path cli --features gpu     # with GPU

---

Quick start

# Genomics + transcriptomics + epigenomics — HTML report + JSON
multiomics \
  --genomics sample.vcf \
  --transcriptomics expression.tsv \
  --epigenomics methylation.bed \
  --output ./results

# JSON only (no TUI, no HTML) — good for automated pipelines
multiomics \
  --genomics sample.vcf \
  --transcriptomics expression.tsv \
  --epigenomics methylation.bed \
  --json \
  --output ./results

# Cancer sample with WES TMB and reference-guided HRD
multiomics \
  --genomics tumor.vcf \
  --transcriptomics rna.tsv \
  --epigenomics wgbs.bed \
  --reference GRCh38.fasta \
  --tmb-genome-mb 35 \
  --output ./cancer_results

# Single-cell RNA with GPU UMAP
multiomics \
  --scrna /path/to/10x_mex_dir \
  --output ./sc_results

# Proteomics database search at 1% FDR
multiomics \
  --proteomics run1.mzML run2.mzML \
  --fasta human_proteome.fasta \
  --proteomics-fdr 0.01 \
  --output ./prot_results

# Full multi-omics tumor run
multiomics \
  --genomics tumor.vcf \
  --transcriptomics rna.tsv \
  --epigenomics wgbs.bed \
  --reference GRCh38.fasta \
  --atac peaks.narrowPeak \
  --cnv cnv.vcf \
  --scrna mex/ \
  --proteomics ms1.mzML ms2.mzML \
  --fasta proteome.fasta \
  --preset cancer \
  --output ./full_results

---

Cancer workflow

A typical somatic tumor analysis:

multiomics \
  --genomics somatic_calls.vcf \
  --transcriptomics tumor_rna.tsv \
  --epigenomics wgbs.bed \
  --reference GRCh38.fasta \
  --preset cancer \
  --output ./tumor_report

The HTML report will include:

• FDA biomarker cards — TMB and MSI with eligibility assessment for pembrolizumab
• COSMIC mutational signatures — dominant SBS signatures with etiology
• Immune evasion score — checkpoint gene panel (CD274/PD-L1, CTLA4, LAG3, TIM-3, TIGIT, B2M)
• Polygenic risk scores — germline risk across 6 cancer types from GWAS catalog
• Tumor purity — VAF + methylation cross-validated estimate
• Kataegis loci — hypermutation foci consistent with APOBEC/AID activity
• HRD score — indel spectrum suggestive of BRCA1/2 deficiency
• LOH map — per-chromosome allele imbalance

---

Input formats

Flag	Format
--genomics	VCF 4.x (gzipped or plain). QUAL, INFO/AF, and gene name from INFO/ANN are used when present.
--transcriptomics	Tab-separated matrix: first row = sample names, first column = gene ID, values = TPM (or raw counts with --raw-counts).
--epigenomics	BED with columns: chrom, start, end, name, score, strand, methylation% (ENCODE bisulfite). A simpler 4-column BED (chrom, start, end, methylation%) is also accepted.
--reference	Reference FASTA (GRCh38 recommended). Enables reference-guided HRD microhomology scoring.
--atac	ENCODE narrowPeak (BED6+4). Requires --features atac.
--cnv	VCF with CN=<int> in INFO. Requires --features cnv.
--scrna	10x Genomics MEX directory: matrix.mtx.gz, barcodes.tsv.gz, features.tsv.gz.
--proteomics	One or more mzML files.
--proteomics-dir	Directory of *.mzML files — scanned automatically.
--fasta	Protein FASTA for proteomics search.
--bam	BAM for gene quantification (requires --gtf).
--gtf	GTF or GFF3 annotation.
--fastq	FASTQ for read-level QC.
--gmt	Gene-set file in GMT format (name, description, genes…).

---

Output files

All files are written to --output (default: ./multiomics_out).

File	Description
report.html	Self-contained HTML report — inline SVG charts, no external dependencies, opens in any browser.
multiqc_multiomics.json	MultiQC-compatible JSON with all summary statistics, biomarker results, and integration outputs.

---

All CLI flags

Primary inputs

Flag	Description
--genomics FILE	VCF for variant analysis
--transcriptomics FILE	Expression matrix TSV
--epigenomics FILE	Bisulfite BED for methylation
--reference FILE	Reference FASTA (enables reference-guided HRD microhomology)
--atac FILE	ATAC-seq narrowPeak
--cnv FILE	VCF with CN field
--fastq FILE	FASTQ for QC
--scrna DIR	10x MEX directory for single-cell analysis
--bam FILE	BAM for gene quantification
--gtf FILE	GTF/GFF3 annotation

Proteomics

Flag	Default	Description
--proteomics FILE…	—	One or more mzML files
--proteomics-dir DIR	—	Directory of mzML files
--fasta FILE	—	Protein database FASTA
--proteomics-fdr N	0.01	FDR threshold (0.01 = 1%)
--phospho-max-sites N	0	Max phospho sites per peptide; 0 disables phosphoproteomics

Somatic variant calling

Flag	Default	Description
--tumor-bam FILE	—	Tumor BAM (requires --normal-bam)
--normal-bam FILE	—	Matched normal BAM
--somatic-min-lod N	6.3	Minimum log-odds score for PASS calls

TMB and cancer biomarkers

Flag	Default	Description
--tmb-genome-mb N	auto	Effective genome size in Mb for TMB. Auto-detected: WGS ≈ 2800, WES ≈ 35.

Comparison mode (tumor vs. normal / treatment vs. control)

Flag	Description
--compare-genomics FILE	Control VCF (enables comparison mode)
--compare-transcriptomics FILE	Control expression TSV
--compare-epigenomics FILE	Control methylation BED
--compare-atac FILE	Control ATAC narrowPeak

Single-cell options

Flag	Default	Description
--umap-neighbors N	15	Number of UMAP neighbours
--no-umap	off	Skip UMAP embedding (clustering still runs)

Output and performance

Flag	Default	Description
--output DIR	./multiomics_out	Output directory (created if absent)
--threads N	all cores	Worker thread count
--json	off	JSON output only — no TUI, no HTML
--no-ml	off	Skip PCA and cross-modality correlation
--no-gpu	off	Disable GPU acceleration; use CPU for UMAP
--raw-counts	off	Treat --transcriptomics as raw counts and apply DESeq2 normalisation

Skip flags — disable individual modules

Flag	What it skips
--skip-genomics	Variant analysis
--skip-transcriptomics	Expression analysis
--skip-epigenomics	Methylation analysis
--skip-proteomics	Proteomics database search
--skip-scrna	Single-cell analysis

Presets and configuration

Flag	Description
--preset NAME	Load a predefined threshold set
--list-presets	Print all available presets and exit
--config FILE	TOML configuration file
--dump-config	Print the default configuration as TOML and exit

---

Skip flags in practice

# Provide VCF but skip genomics (run only transcriptomics + epigenomics)
multiomics \
  --genomics mutations.vcf \
  --transcriptomics rna.tsv \
  --epigenomics wgbs.bed \
  --skip-genomics

# Full pipeline but skip the proteomics database search
multiomics \
  --genomics tumor.vcf \
  --transcriptomics rna.tsv \
  --epigenomics wgbs.bed \
  --proteomics mass_spec.mzML \
  --fasta proteome.fasta \
  --skip-proteomics

# Single-cell without UMAP (faster QC + normalisation only)
multiomics --scrna mex/ --no-umap

# Force CPU UMAP even when GPU is available
multiomics --scrna mex/ --no-gpu

# Skip ML integration layer (no PCA, no correlation matrix)
multiomics \
  --genomics sample.vcf \
  --transcriptomics rna.tsv \
  --epigenomics wgbs.bed \
  --no-ml

---

Presets

Presets adjust analysis thresholds for common study types without editing a config file.

Preset	Use case
cancer	Somatic mutation analysis; lower Ti/Tv warning threshold
plant	Plant genome; adjusted GC and methylation norms
rna-seq	Expression-focused; DE-centric insights
wgbs	Whole-genome bisulfite; strict CpG island parameters
atac	Chromatin accessibility; ATAC signal thresholds
clinical	Clinical WGS; conservative QUAL and FDR cutoffs

multiomics --genomics tumor.vcf --preset cancer --output ./results
multiomics --list-presets

---

Configuration file

The full set of thresholds and performance knobs can be saved and edited as TOML:

multiomics --dump-config > my_config.toml
# edit my_config.toml to taste
multiomics --config my_config.toml --genomics sample.vcf ...

---

GPU acceleration details

When compiled with --features gpu, UMAP uses a WebGPU compute shader for the k-NN step.

Algorithm — GPU k-selection: one GPU thread per cell. Each thread keeps a private max-heap of size k (max 64), scans every other cell, and writes only n×k indices and n×k distances. No n×n distance matrix is ever allocated or transferred.

• Readback size: n × k × 8 bytes — about 12 MB for 100 000 cells at k=15
• Falls back to CPU automatically when n < 5 000 or no GPU adapter is found

Platforms: Vulkan (Linux, Windows), Metal (macOS / Apple Silicon), DX12 (Windows). No CUDA dependency.

Benchmark on RTX 4050 (6 GB VRAM):

Cells	CPU	GPU (k-selection)
2 000	~730 ms	CPU fallback
5 000	~1.8 s	~1.6 s
10 000	~7 s	~5 s
20 000	~30 s	~18 s
50 000	~200 s	~90 s

---

Feature flags at build time

Cargo feature	What it adds
(none / default)	Genomics, transcriptomics, epigenomics, ATAC-seq, CNV
gpu	GPU-accelerated UMAP (wgpu / Vulkan / Metal / DX12)
longread	Long-read (PacBio / Nanopore) epigenomics parsing
full	ATAC + CNV + longread
full,gpu	Everything

---

Scientific references

Feature	Reference
TMB	Chalmers et al. 2017 (Genome Medicine); FDA pembrolizumab approval 2020
MSI	Bonneville et al. 2017 (JCO Precision Oncology); Cortes-Ciriano et al. 2017 (Nature Comms); FDA approval 2017
COSMIC mutational signatures	Alexandrov et al. 2020 (Nature); COSMIC v3.3
Immune evasion score	Chen & Mellman 2017 (Nature); Ribas & Wolchok 2018 (Science)
PRS	NHGRI-EBI GWAS Catalog; variants with p < 5×10⁻⁸
Kataegis	Alexandrov et al. 2013 (Nature)
HRD indel score	Watkins et al. 2020 (Nature Genetics); Chan et al. 2015 (Nature Genetics)
Tumor purity	Carter et al. 2012 (Nature Biotechnology); ABSOLUTE
Horvath epigenetic clock	Horvath 2013 (Genome Biology); 353 CpG sites, hg19
MOFA+	Argelaguet et al. 2018 (Molecular Systems Biology)
GSEA	Subramanian et al. 2005 (PNAS)
CpG islands	Gardiner-Garden & Frommer 1987 (Journal of Molecular Biology)

---

Development

# Verify all crates compile
cargo check --workspace

# Lint — zero warnings policy
cargo clippy --workspace -- -D warnings

# Tests
cargo test --workspace

# UMAP benchmarks (CPU vs GPU)
cd scrna_core
cargo bench                    # CPU only
cargo bench --features gpu     # CPU + GPU comparison

---

License

Apache-2.0 — see LICENSE.