Biology subjects

Vaz, E.

Publications and source records attributed to Vaz, E..

4 recordsLinked to original sources

bioRxiv2026

Synthetic MPRA perturbation supervision improves variant-effect prediction in sequence-to-function models

Deep learning sequence-to-function (S2F) models learn to predict functional readouts, such as gene expression, from genetic sequence. However, these models are trained predominantly on the reference genome and struggle to accurately predict the effect of genetic variation on gene expression. Massively parallel reporter assays (MPRAs) explore the direct impact of genetic variation on gene expression, but saturation-mutagenesis MPRAs that measure variant effects directly remain limited, thereby restricting their use for fine-tuning S2F models. In this work, we investigate fine-tuning Borzoi, a state-of-the-art S2F model, on model-generated synthetic MPRA variant-effect data as a scalable alternative to overcome limits on the availability of supervised training data for genetic variation. First, we trained an ensemble of 16 models on 110,359 K562 MPRA measurements. Then, we distilled their predictions into a single student model using MPRA elements associated with 1,000 genes. The distilled model was subsequently used to generate synthetic variant-effect labels across increasing numbers of genes (50 to 1,000), with 600 variants per gene, to fine-tune Borzoi. Performance improved as we included more genes in training. In data-limited settings, training on more genes with fewer variants per gene performed better than training on fewer genes with more variants per gene, given the same number of total variants. Compared to pre-trained Borzoi, Borzoi fine-tuned on 600,000 synthetic samples across 1,000 genes increased Pearson correlation from 0.28 to 0.54 on CAGI5 promoter variants and from 0.39 to 0.46 on 903 GTEx fine-mapped whole-blood expression quantitative trait loci (eQTLs). Importantly, these improvements were achieved while preserving the pre-trained performance across all 7,611 output tracks. Furthermore, fine-tuned Borzoi recovered regulatory motifs identified by the MPRA model in genes unseen during training, suggesting that learned regulatory features can generalize across loci. Together, these results provide a proof of concept that model-generated synthetic supervision can improve variant effect prediction in S2F models while preserving their original pre-trained capabilities, offering a scalable strategy toward capturing the impact of genetic variation on gene regulation.

genomics↗

bioRxiv2025

Fine-tuning sequence to function deep learning models on large-scale proteomic data improves the accuracy of variant effect prediction

Fine-tuning sequence to function models has shown promise for variant effect prediction, but accuracy and generalization to unseen genes and unseen individuals remains a standing challenge. We fine-tuned Borzoi on 54,219 individuals and 2,923 circulating plasma proteins from the UK Biobank Plasma Proteomic Project. Across 150 single-gene models where the genes had a range of cis-heritability we observed that the fine-tuned Borzoi model improved variant effect prediction for 86% of the genes compared to an Elastic Net baseline model. We demonstrated that the improved prediction stems from increased sample size which provides tremendous amounts of rare genetic variants (MAF < 0.01) to the training data. Masking rare and uncommon variants nullified improved performance of fine-tuned Borzoi and we showed that fine-tuned Borzoi highly weights rare variants (MAF < 0.01) while the Elastic Net model highly weights common variants (MAF > 0.05) that are enriched for regulatory regions. We evaluated the generalizability of our model on a fine-tuned Borzoi model trained jointly on varying numbers of genes and observed that these models consistently outperform the pre-trained Borzoi model, the single-gene models yield more accurate results. Together this work demonstrates the importance of including larger sample sizes and rare variants in sequence to function models for variant effect prediction and demonstrates feasibility that these models are capable of highly accurate variant effect prediction.

genetics↗

bioRxiv2023

Engineering self-propelled tumor-infiltrating CAR T cells using synthetic velocity receptors

Chimeric antigen receptor (CAR) T cells express antigen-specific synthetic receptors, which upon binding to cancer cells, elicit T cell anti-tumor responses. CAR T cell therapy has enjoyed success in the clinic for hematological cancer indications, giving rise to decade-long remissions in some cases. However, CAR T therapy for patients with solid tumors has not seen similar success. Solid tumors constitute 90% of adult human cancers, representing an enormous unmet clinical need. Current approaches do not solve the central problem of limited ability of therapeutic cells to migrate through the stromal matrix. We discover that T cells at low and high density display low- and high-migration phenotypes, respectively. The highly migratory phenotype is mediated by a paracrine pathway from a group of self-produced cytokines that include IL5, TNF, IFN{gamma}, and IL8. We exploit this finding to "lock-in" a highly migratory phenotype by developing and expressing receptors, which we call velocity receptors (VRs). VRs target these cytokines and signal through these cytokines cognate receptors to increase T cell motility and infiltrate lung, ovarian, and pancreatic tumors in large numbers and at doses for which control CAR T cells remain confined to the tumor periphery. In contrast to CAR therapy alone, VR-CAR T cells significantly attenuate tumor growth and extend overall survival. This work suggests that approaches to the design of immune cell receptors that focus on migration signaling will help current and future CAR cellular therapies to infiltrate deep into solid tumors.

immunology↗

bioRxiv2023

Three-dimensional assessments are necessary to determine the true spatial tissue composition of diseased tissues

Methods for spatially resolved cellular profiling using thinly cut sections have enabled in-depth quantitative tissue mapping to study inter-sample and intra-sample differences in normal human anatomy and disease onset and progression. These methods often profile extremely limited regions, which may impact the evaluation of heterogeneity due to tissue sub-sampling. Here, we applied CODA, a deep learning-based tissue mapping platform, to reconstruct the three-dimensional (3D) microanatomy of grossly normal and cancer-containing human pancreas biospecimens obtained from individuals who underwent pancreatic resection. To compare inter-and intra-sample heterogeneity, we assessed bulk and spatially resolved tissue composition in a cohort of two-dimensional (2D) whole slide images (WSIs) and a cohort of thick slabs of pancreas tissue that were digitally reconstructed in 3D from serial sections. To demonstrate the marked under sampling of 2D assessments, we simulated the number of WSIs and tissue microarrays (TMAs) necessary to represent the compositional heterogeneity of 3D data within 10% error to reveal that tens of WSIs and hundreds of TMA cores are sometimes needed. We show that spatial correlation of different pancreatic structures decay significantly within a span of microns, demonstrating that 2D histological sections may not be representative of their neighboring tissues. In sum, we demonstrate that 3D assessments are necessary to accurately assess tissue composition in normal and abnormal specimens and in order to accurately determine neoplastic content. These results emphasize the importance of intra-sample heterogeneity in tissue mapping efforts.

cancer biology↗