We have moved beyond the genome-wide association study (GWAS) paradigm entirely. While the field debates common versus rare variants, we have found that the most valuable information is hidden in what everyone discards.
Our patent-pending computational platform extracts molecular subtypes from genomic signals that conventional pipelines discard, applying machine learning and family-based inheritance analysis to identify biologically distinct disease subtypes from existing datasets.
Analyze trio and family genomic data to capture inheritance patterns unavailable in case-only analyses.
Apply proprietary algorithms to extract subtype-defining signals from genomic patterns typically filtered out as noise — including probe intensity anomalies, copy number variants, and non-Mendelian inheritance signatures.
Unsupervised learning methods group individuals by molecular signal profiles, revealing biologically coherent subtypes without requiring predefined labels.
Map extracted signals to biological pathways and molecular mechanisms, generating mechanistically interpretable subtype definitions.
Conventional genetic analysis is built around filters designed to reduce noise. Those same filters discard rare variants, structural signals, and inheritance patterns that carry biologically meaningful information.
| Conventional Approaches | Williwaw Platform |
|---|---|
| GWAS identifies common variants of small effect | Detects rare variants from accessible SNP arrays |
| Rare variant detection requires expensive NGS | Extracts signal from probe intensity anomalies |
| Single-sample analysis misses inheritance patterns | Leverages family-based genomic patterns |
| Copy number gains treated as technical artifacts | Copy number gains reveal structural variants |
| Subtypes invisible to standard pipelines | Reveals molecular subtypes with pathway resolution |
Large-scale genetic databases already contain the subtyping information we need — it has been systematically filtered out by conventional pipelines. We do not require new data collection; we extract previously invisible biological signals from existing datasets.
Autism was our initial application because it represents one of the most genetically complex conditions in medicine: highly heritable yet extremely heterogeneous, with no reliable biomarkers and limited treatment options. Our findings are published in Frontiers in Molecular Neuroscience and HGG Advances.
Autism spectrum disorder encompasses the full complexity of human genetic architecture:
Unsupervised machine learning identifies six biologically distinct autism subtypes with different underlying mechanisms (n=1,811).
Each subtype has distinct biological mechanisms, clinical characteristics, and therapeutic implications.
Therapeutic implication: Requires comprehensive, multi-system approach rather than single-pathway intervention.
Therapeutic implication: Harbors variants in glutamate receptors and ion channels with FDA-approved therapies for other conditions — an immediate drug repurposing candidate.
Therapeutic implication: FDA-approved drug targets include cholesterol receptors, GABA-B receptors, and neurotrophin pathways. Anti-inflammatory strategies may be relevant.
Therapeutic implication: Potentially responsive to cholesterol-modulating drugs.
Therapeutic implication: Early motor interventions; occupational and physical therapy focused on motor development.
Therapeutic implication: Strong candidate for SSRI response. May respond to dietary tryptophan modulation.
Neurotransmission, vesicle recycling, and synaptic plasticity
Excitability regulation, E/I balance, action potential generation
Cell-cell contacts, dendritic spine formation, synapse stability
Microtubule organization, axon guidance, neuronal polarity
Intracellular transport, endosomal sorting, membrane dynamics
Synaptic pruning, maternal immune activation, inflammatory responses
DNA methylation, RNA modification, chromatin remodeling
mRNA degradation, translation control, post-transcriptional regulation
Our platform identifies molecular subtypes in any genetically complex condition — including diseases traditionally considered simple Mendelian disorders. Even single-gene diseases show unexpected molecular diversity.
Even single-gene diseases are molecularly heterogeneous. Standard genetic testing identifies the primary mutation but cannot predict severity or treatment response. Characterizing the modifier landscape that determines clinical outcomes is where precision medicine for rare disease begins.
Iron overload disorder
Patients with identical HFE C282Y/C282Y genotypes show 100-fold variation in outcomes — some develop severe cirrhosis by age 40, others remain asymptomatic for life. Penetrance of HFE mutations is approximately 1%, and no clinical factors reliably explain this variation.
Both HFE (iron sensor) and MTF1 (zinc sensor) regulate HAMP expression. Zinc transporter disruption via SLC39A12 compounds the dysregulation, representing a previously undescribed iron-zinc epistatic interaction in hemochromatosis. Both NMI loci are eQTLs regulating gene expression.
Iron-zinc regulatory pathway showing multiple hemochromatosis types. HFE (Type 1) requires modifier genes for severe phenotype. Rare variants in SLC39A12 (zinc transporter) and MTF1 (zinc-responsive transcription factor) regulate HAMP expression, creating epistatic interaction with HFE mutations. Other hemochromatosis genes shown: HJV (Type 2A), HAMP (Type 2B), TFR2 (Type 3), SLC40A (Type 4), FTH1 (Type 5), plus iron metabolism modifiers (TMPRSS6, BMP2, BMP6).
Severe progressive neurodegenerative ataxia
The same NPC1 mutations cause infantile vs. adult-onset disease (decades apart), with variable progression from rapid neurodegeneration to slow decline. The prevailing cholesterol-storage model has driven therapies with limited efficacy. Our analysis points to iron-dependent ferroptosis as the primary driver, opening different therapeutic avenues.
Complete ferroptosis pathway with modifier genes and FDA-approved drug targets. NPC1 oxidation triggers iron accumulation → lipid peroxidation → ferroptotic cell death. Modifier genes affect the cholesterol pathway (NPC1, NPC2, MALL, COX10, GCLC), sphingomyelin pathway (SMPD1, AGMO, PAH, NPC2), and lipid peroxidase pathway (ALOX5, AGMO, PSMB8). NRF2 activation by omaveloxolone targets multiple pathway components including ALOX5, PSMB8, GCLC, STING, and SLC40A1.
| Approach | Limitation |
|---|---|
| Standard genetic testing | Identifies the primary mutation but ignores modifier variants that determine severity |
| Whole genome sequencing | Generates millions of variants but lacks the framework to identify which modifiers matter |
| Our platform | Extracts subtype-defining signals from family data and integrates regulatory variants to reveal functional subtypes |
Our molecular subtyping platform enables precision medicine across diverse therapeutic areas and stakeholders.
The methods validated in autism apply to any complex condition with a genetic component and available family-based or population genomic data. Potential applications include: psychiatric disorders (schizophrenia, bipolar disorder, ADHD), neurodevelopmental conditions (intellectual disability, epilepsy), cancer molecular subtypes, cardiovascular disease stratification, autoimmune conditions, rare disease characterization, pharmacogenomic response prediction, and complex trait architecture.
Short, evidence-driven perspectives on what's working — and what's not — in precision medicine.
February 2026
Precision medicine promises healthcare tailored to the individual. But there's an uncomfortable truth: the "precision" part is still struggling — especially in genetics.
Polygenic risk scores (PRS) are often presented as a breakthrough. They combine thousands of tiny genetic signals to predict disease risk. Yet for many real-world clinical decisions, the improvements are small, inconsistent across populations, and rarely definitive for individual care.
Even in human height, one of the most heritable and best-studied traits, the largest genetic studies show a ceiling. Tens of thousands of common variants explain only part of the biology, leaving substantial variation outside the model.
Instead of closing those biological gaps, the industry is pivoting toward:
These tools are valuable — but they are increasingly compensating for a core limitation: our genetic measurements remain incomplete.
PRS rely mostly on common SNPs, which are easy to measure but capture only a slice of genomic complexity. They largely miss structural variation, gene interactions, and regulatory effects that often drive real disease mechanisms.
The industry is building increasingly sophisticated analytics on top of a simplified view of genetics.
That's not precision. That's approximation.
If precision medicine is truly about the individual, we can't keep skimming the surface of the one dataset that is fundamentally individual: the genome itself. Until we confront that, AI will keep getting smarter while biology remains under-measured.
We are seeking strategic partners to apply our molecular subtyping platform across therapeutic areas.
We are open to collaborations with academic medical centers, pharmaceutical and biotechnology companies, genetic testing providers, and rare disease foundations. Our platform can be applied to existing datasets, making partnerships straightforward to initiate.
Get In TouchDr. Garvin is a molecular geneticist with over 20 years of experience in precision medicine and computational biology. He holds a Ph.D. from the University of Alaska Fairbanks and a Certificate in Personalized & Genomic Medicine from the University of Colorado Denver. His career spans Oak Ridge National Laboratory, Oregon State University, and biotechnology companies including Tularik (Amgen) and CV Therapeutics (Gilead).
His research focuses on identifying the genetic basis of complex diseases through novel computational approaches. With over 37 peer-reviewed publications in journals including Genome Biology, eLife, and PLoS Genetics, and over $5 million in NIH and DOE funding, Dr. Garvin has established a record in precision medicine research.
His COVID-19 research became the 2nd most viewed article in eLife history with over 154,000 views. His autism subtyping work is published in Frontiers in Molecular Neuroscience and HGG Advances. Dr. Garvin founded Williwaw Biosciences after discovering that family-based genomic data contains rich subtyping information systematically discarded by conventional analysis pipelines.
Ben Garvin brings extensive experience in technology commercialization and strategic partnerships, with a track record of selling advanced algorithmic and data products to large enterprises.
As Chief Business Officer, Ben leads business development, strategic partnerships, and commercialization efforts for Williwaw Biosciences.
Interested in applying our platform to your therapeutic area or exploring partnership opportunities?
For clinical partnerships, licensing inquiries, or investment opportunities, please reach out directly.