🛑TLDR: It’s time to stop asking “Is this variant pathogenic?” — and instead ask “What’s its disease risk — and for whom?” Read the full perspective: https://t.co/cxrNgQPnvq 👈 #Genomics #medicine #PrecisionMedicine

Massive gratitude to all coauthors, institutions, and our journal partners: @KuanlinHuang @WendyKChung @DrDanielJordan @Dr_J_Eggington @DoGenetics @IcahnMountSinai @NatureGenet @SpringerNature Thank you all for the collaboration & support 🙏

Imagine a variant report that reads: “This variant carries 12% lifetime risk in women, 4% in men.” Not just “Pathogenic.” This is the precision-medicine future we advocate for 💪 And AI can help us get there (see our recent Science study: https://t.co/VKbAC7iyrz 👈)

This debate has deep roots as old as Mendel: Mendelian discrete vs. biometric continuous inheritance 🫛 Our approach? Bridge the divide by combining categorical calls from detailed clinical studies with probabilistic risk from large-scale population data 🧬📈

ClinVar is a commonly used resource but missing key pieces: • Real-world penetrance estimates (this is finally changing as of 2025) • Age-/sex-/ancestry-stratified risk • Systematic integration from population cohorts We propose linking a penetrance repository into ClinVar 🔗

Not all methods to assess disease-risk variants are created equal: ⚠️ ACMG/ClinVar: simple but static ⚠️ Penetrance: quantitative but at risk for being underpowered ✅ Bayesian integration: combines evidence, updates with data

We need to move from labels → quantitative risk 📊 • What proportion of carriers develop disease? • By what age? • In which ancestries or sexes? • In what environmental context?

E.g., a BRCA2 variant previously thought to be “pathogenic” was observed at a nontrivial frequency in healthy cohorts. ➡️ It was reclassified accordingly ➡️ Clinical management for those patients changed.

Clinicians acting on “pathogenic” calls of variants lack large truth sets to accurately gauge disease risk 📈 If only a small % of carriers ever show disease, those labels may mislead… This isn’t theoretical—we see it across genes and disease domains in several studies.

For decades, variants have been roughly boxed into: ✅ Pathogenic ⚠️ Uncertain Significance (VUS) ❌ Benign But these binary labels hide critical nuance—they don’t tell patients how likely disease is, or for whom we should be more worried about…

🚨“Pathogenic” genetic variant does NOT = guaranteed disease 🚨 Yet, doctors and patients are still left to decide what to do with the label… In our new @NatureGenet perspective, we explain why this is and how we can improve genetic risk interpretation🧵

The O’Neal lab at @JohnsHopkinsSPH is hiring a Lab Manager to work on ticks and tick-borne diseases. You can apply here:

TLDR Post-Labor Day inspiration to start your week: AI + EHR data may finally answer the question patients ask most… 👉 What are the chances this DNA variant actually make me sick? Read the full paper in @ScienceMagazine: https://t.co/VKbAC7iyrz #PrecisionMedicine #Genomics #AI

It took a village! Forever grateful to @DoGenetics Lab @IcahnMountSinai including @DrDanielJordan @OmegaPetrazzini and collaborators @girish_nadkarni @JudyCho7 @WendyKChung. No better way to kick off internal medicine residency @BayviewMedicine @HopkinsBayview @JohnsHopkins 💪

🧬ML penetrance = a scalable, data-driven way to quantify disease risk. It 1️⃣ Improves variant interpretation 2️⃣ Reduces patient uncertainty 3️⃣ Bridges genomics with real-world medicine!

As Science put it: pathogenicity and penetrance are “two sides of the same coin.” ML penetrance brings the clinical spectrum into the equation — complementing lab predictions like AlphaMissense ⚖️ Check out the full perspective by @_amelie_rocks 👉 https://t.co/e8iRwH6eh6

This blueprint could help: • Genetic counselors provide quantitative risk “60%” “10%”, not just labels “pathogenic” “benign”, to patients 📊 • Clinicians guide earlier screening/prevention for high risk variants 🩻 • Researchers prioritize variants for functional follow-up 🧫

One of my favorite results! For patients with a “VUS” (variant of uncertain significance), ML penetrance offers clarity 🔮 Example: PKD VUS with high ML penetrance showed kidney function decline, while low-penetrant ones looked like controls…

Traditional penetrance estimates = typically binary: either you have the disease or not. For rare variants with only 1–2 carriers, that means crude values (0, 0.5, or 1) 🔨 ML penetrance uses continuous disease scores, giving refined estimates for every individual 🔪

ML penetrance wasn’t just theoretical — it correlated with: 📈 Clinical outcomes (e.g. ESRD in PKD, heart failure in HCM) 🧪 Functional assays (LDLR cholesterol uptake, BRCA1 repair👇 see below, KCNQ1 electrophysiology) Bridging two different but crucial types of phenotypes…