RT @raffcolo: Is there a synergistic effect between enfortumab vedotin (EV) and pembrolizumab (P) for untreated advanced urothelial carcino….

These findings suggest that distinct patient subgroups respond to EV vs pembro, as opposed to an emergent synergistic effect with combination EV+P. Important implications for drug development in this setting! @PennCancer (4/4).

Blazing-fast image creation – using just your voice. Try Grok Imagine.

In an exploratory analysis, we also evaluated patients with PD-L1 CPS >10, who are more likely to respond to pembro. In this biomarker-selected group, observed PFS for EV-302 was again well-explained by independent activity of EV and pembro. (3/4)

Inspired by pioneering work from @ac_palmer, we calculated the predicted PFS curve for combo EV+P, assuming independent activity of EV and pembro. Amazingly, predicted PFS for EV+P was essentially indistinguishable from what was observed in the EV-302 trial. @tompowles1 (2/4)

Enfortumab vedotin + pembrolizumab (EV+P) has revolutionized how we treat urothelial cancer. @Ron_cology and I were curious: is the efficacy of EV+P better explained by synergistic or independent drug action? New in @UrolOncol: (1/4).

RT @NatureBiotech: Multiplexed inhibition of immunosuppressive genes with Cas13d for combinatorial cancer immunotherapy .

RT @ravi_b_parikh: New @Nature_NPJ paper by @ScienceChow co-mentored w/ @KLNathanson @BasserBRCA on reliability of #deeplearning to predict….

RT @BiologyAIDaily: Phenotypic Evaluation of Deep Learning Models for Classifying Germline Variant Pathogenicity @Nature_NPJ . • This study….

Thanks for reading! Open-access at This study would not be possible without all the @uk_biobank participants. Many thanks to @KLNathanson & @ravi_b_parikh for their mentorship! @PennCancer @PennMedicine @PC3Innovation @PennIMResidents (8/8).

The notable exception: PALB2. Considering the sparse/conflicting ClinVar annotations for PALB2, this represents a concrete example where current #DeepLearning models could already inform variant classification and thus impact clinical decision making. (7/8).

However, we also noted smaller effect sizes vs ClinVar, suggesting dilution of discriminative power. When we specifically tasked the models with classifying VUS, they largely failed to identify VUS that confer increased #BreastCancer risk. (6/8)

Our approach: let's go straight to the actual disease phenotype that we care about -- in this case, #BreastCancer risk. Leveraging the @uk_biobank, we saw that model-based predictions for 3/5 genes were significantly associated with cancer risk. (5/8)

But, the goal of these models is to classify variants of uncertain significance (VUS) that are *not* well-annotated in ClinVar. Without a gold-standard reference for VUS, how can we meaningfully test model performance on these variants? (4/8).

In line with prior studies, the deep learning models generally recapitulated "gold-standard" pathogenic vs benign variant classifications from ClinVar. (3/8)

We applied 3 state-of-the-art models (AlphaMissense @GoogleDeepMind, ESM1b, and EVE) to classify germline variants in over 450,000 @uk_biobank participants, focusing on hereditary #BreastCancer genes (eg BRCA1, BRCA2). (2/8)

#DeepLearning models have been developed to predict missense variant pathogenicity -- but how well do these models perform in a real-world clinical setting? Thrilled to share our latest in npj Precision Oncology! @Nature_NPJ | (1/8).

RT @medrxivpreprint: Real-world evaluation of deep learning algorithms to classify functional pathogenic germline variants .

That @ericsongg could pull this off *WHILE IN RESIDENCY* is absolutely bonkers to me. 10/10 would let him inject AAVs into my eyes any day.

RT @ericsongg: Excited to share our group's first manuscript describing new anatomical connections that allow the eyes to act as an "immun….

RT @sidichen: Our work at Nature Immunology on a new way of cell therapy engineering. Proud of 1st author Ariel & team. Special Tx to Dr.….