Abstract CRISPR-based genetic perturbation screens paired with single-cell transcriptomic readouts (Perturb-seq) offer a powerful tool for interrogating biological systems. Yet the resulting datasets are heterogeneous—particularly in vivo—and currently used cell-level perturbation labels reflect only CRISPR guide RNA exposure rather than perturbation state; further, many perturbations have a minimal effect on gene expression. For perturbations that do alter the transcriptomic state of cells, intracellular guide RNA abundance exhibits a dose-response association with perturbation efficacy. We combine (i) per-perturbation, expression-only classifiers trained with non-negative negative–unlabeled (nnNU) risk to yield calibrated scores reflecting the perturbation state of single cells and (ii) a monotone guide abundance prior to yield a per-cell pseudo-posterior that supports both assignment of perturbation probability and selection of affected gene features. To obtain a low-dimensional representation that allows for the accurate reconstruction of gene-level marginals for counterfactual decoding, we train an autoencoder with a quantile–hurdle reconstruction loss and feature-weighted emphasis on perturbation-affected genes. The result is a perturbation-aware latent embedding amenable to downstream trajectory modeling (e.g., optimal transport or flow matching) and a principled probability of perturbation for each non-control cell derived jointly from its guide counts and transcriptome.

Authors Florian Hugi, Tanmay Tanna, Randall J. Platt, Gunnar Rätsch

Submitted NeurIPS 2025 AI4D3

Link DOI

Abstract Gene therapies are emerging as a new treatment modality. Due to their novelty, general pharmacological properties have yet to be established. For example, the translation from animal models to humans for first-in-human dose selection and the dose-exposure relationship remain poorly characterized. A mechanistic and quantitative framework would improve preclinical program design, enable more robust first-in-human dose predictions, and support more rigorous dose adjustments during clinical development. This study establishes a semimechanistic mathematical model for aflibercept expression and pharmacokinetics (PK) following intravitreal (IVT) ADVM-022 administration in monkeys and humans, drawing on the preclinical and clinical data presently available. ADVM-022 is an AAV2.7m8-based viral vector that delivers the gene encoding aflibercept, an antivascular endothelial growth factor (VEGF) fusion protein. It was developed as a gene therapy for treating wet age-related macular degeneration (wAMD) and is administered through a single IVT injection. The proposed model incorporates established ocular PK for intravitreally administered proteins, along with an expression component that links AAV dose to aflibercept production. Based on pooled PK data from monkey studies, the model suggests that transduction occurs not only in the retina but also in other ocular tissues bordering the vitreous, contributing to the observed intraocular aflibercept levels. Increasing doses within the lower range of preclinical studies (3 × 1010–2 × 1013 vg/eye) lead to increased transduction and expression, plateauing at upper limits of approximately 12.7 μg/day·cm3 for the retina, and 0.785 μg/day for extra-retinal tissues at higher doses. Assuming similar transduction efficiency between humans and monkeys, with adjustments for anatomical differences, the model provided predictions of ocular aflibercept concentrations that aligned with observations from the two dose groups in the phase 1 OPTIC clinical trial, supporting the utility of this approach.

Authors Florian Hugi, Jannik Vollmer, Lionel Renaud, Matthias Machacek

Submitted Molecular Pharmaceutics

Link DOI