Biology subjects

Crawford, L.

Publications and source records attributed to Crawford, L..

At least 19 recordsLinked to original sources

bioRxiv2026

Single-cell perturbation prediction near experimental reproducibility with GENESIS

Single-cell perturbation atlases now measure cellular responses across millions of cells, but the number of possible combinations of biological contexts, perturbations, doses and other experimental variables make exhaustive measurements infeasible. Here, we present GENESIS (Generative ENErgy-based Single-cell Intervention Simulator), a generative framework that predicts single-cell response distributions directly from the covariates defining an experimental condition. GENESIS uses a lightweight neural network trained with the energy score, a strictly proper scoring rule for distributions. We introduce an experiment-calibrated evaluation framework in which prediction error is interpreted relative to the reproducibility of repeated measurements of the same condition. Across the Tahoe-100M small-molecule atlas and a large PBMC cytokine atlas, GENESIS significantly outperforms existing generative models and statistical baselines and approaches experimental reproducibility on independently repeated conditions. Importantly, GENESIS remains accurate when trained on only a fraction of the available experimental conditions or cells, and generated profiles recover experimentally observed context-dependent transcriptional programs. At comparable training-cell budgets, GENESIS achieves more accurate predictions when measurements span more experimental conditions rather than profiling fewer conditions more deeply. Together, these findings suggest a complementary strategy for perturbation studies: broader, lower-depth measurements can expand experimental coverage, while GENESIS augments sparsely sampled populations and predicts responses in unmeasured conditions.

bioinformatics↗

bioRxiv2026

Protein language models and the long tail of functional diversity

Protein language model performance on downstream tasks depends on the pretraining data, motivating recent efforts to combine genomic- and metagenomic-derived protein sequences into large-scale atlases. Because these datasets are highly redundant, sequences are typically clustered by similarity and sampled during training. Sequences that do not belong to any cluster, known as "singletons", are typically excluded from training and evaluation because they are considered to be artifacts. However, singletons represent the long tail of functional diversity and are abundant in many large-scale atlases: nearly 43% of the 3.34 billion sequences in the joint genomic-metagenomic dataset GigaRef are singletons. Here, we characterize singletons derived from UniRef and GigaRef by assessing whether clustering missed homologs, how much their exclusion affects protein language model (PLM) training, and which biological domains they contain. We find that many GigaRef singletons belong to a cluster under alternative parameter settings, suggesting that genomic and metagenomic datasets may require dataset-specific clustering configurations. We also show that singletons share mutual information with clustered sequences, making them learnable by PLMs and useful for training. Finally, metagenomic singletons carry denser, more diverse domain content than clustered sequences, including domain-level homology that sequence-identity clustering misses. Together, these results support including singletons in PLM training and call for closer examination of data curation in large-scale integrated sequence atlases.

bioinformatics↗

bioRxiv2026

Evolutionary conditioning enables guided generation of functionally diverse enhancers

Deep learning has been instrumental in our understanding of how enhancers encode regulatory information in their DNA sequence and has demonstrated preliminary success with enhancer design. However, the prevailing approach for enhancer design, cell type label conditioning, depends on labeled data from massively parallel reporter assays, which only exists for a handful of cell types. We propose EnhancAR, an autoregressive model trained on sets of unaligned homologous enhancer sequences to learn the function of the enhancer conserved over evolution and generate sequences that resemble real homologs. By training EnhancAR on 1.7 million human enhancer homolog sets spanning 1,888 cell types, EnhancAR generates enhancers for a variety of contexts without being conditioned on a cell type label. We computationally validate that when conditioned on a set of enhancer homologs, EnhancAR generates novel and diverse sequences that preserve the functional properties of the homologs. By prompting EnhancAR with homologs for existing cell type specific enhancers, we design enhancers with similar predicted cell type specificity. We further demonstrate that when trained on length sorted homologs, EnhancAR can design enhancers shorter than the conditioning homologs that preserve the predicted activity. In summary, we find that leveraging evolutionary information in enhancer homologs enables a more flexible and general paradigm for designing enhancers with specific functions.

bioengineering↗

bioRxiv2026

Reducing the Foreign Body Reaction to Neuronal Implants in the Central Nervous System with Porous Precision-templated, Mechanically Compliant Hydrogel Scaffolds

Central nervous system diseases and injuries might be treated by implanted devices, tissue regenerative scaffolds, or drug delivery platforms. However, inflammatory CNS responses limit these interventions and may worsen outcomes following damage to the CNS. Via the foreign body reaction, macrophages and glial cells trigger a "glial scar" around implants, reducing device performance, scaffold regenerative ability, or drug delivery potential. Previous studies have shown that stiffness of CNS implants significantly affects glial encapsulation, but few studies have investigated materials that truly match brain tissue stiffness. Porous precision-templated scaffolds with uniform, interconnected, 40 {micro}m spherical pores have shown favorable healing outcomes and a reduced FBR in numerous soft and hard tissue applications. To quantify the effects of both hydrogel compliance (stiffness) and pore size on glial encapsulation, we implanted poly(2-hydroxyethyl methacrylate-co-glycerol methacrylate) (pHEMA/GMA) scaffolds of varying stiffness and pore size for 4 weeks in rat brain. We observed reduced astrocyte encapsulation around PTS compared to solid hydrogel rods, reduced pro-inflammatory macrophage polarization for softer hydrogels versus stiffer hydrogels, and the presence of new blood vessels, neuronal markers and neurogenesis within the pores. Utilizing soft, precision-porous hydrogels could provide a strategy for mitigating glial scarring and improving regeneration in implant-based CNS treatments.

bioengineering↗

bioRxiv2026

Beyond alignment: synergistic integration is required for multimodal cell foundation models

The vision of a "virtual cell"--a computational model that simulates biological function across modalities and scales--has become a defining goal in computational biology. While powerful unimodal foundation models exist, the lack of large-scale paired data prohibits the joint training of multimodal approaches. This scarcity favors compositional foundation models (CFMs): architectures that fuse frozen unimodal experts via a learned interface. However, it remains unclear when this multimodal fusion adds task-relevant information beyond the strongest unimodal representation and when it merely aggregates redundant signal. Here, we introduce the Synergistic Information Score (SIS), a metric grounded in partial information decomposition (PID), that quantifies the information gain achievable only through cross-modal interactions. Extending theoretical results from self-supervised learning, we show that standard alignment-based fusion objectives on frozen encoders inherently collapse to detecting linear redundancies, limiting their ability to capture nonlinear synergistic states. This distinction is directly relevant for tasks aiming to link tissue morphology and gene expression. Benchmarking ten fusion methods on spatial transcriptomics datasets, we use SIS to demonstrate that tasks dominated by linear redundancies are sufficiently served by unimodal baselines, whereas complex niche definitions benefit from synergy-aware integration objectives that enable cross-modal interactions beyond linear alignment. Finally, we perform a scaling analysis which highlights that fine-tuning a dominant unimodal expert is the most sample-efficient path for standard tasks, suggesting that the benefits of multimodal frameworks only emerge when tasks depend on information distributed across modalities. Together, these results establish that building towards a virtual cell will require a fundamental shift from alignment objectives that emphasize shared structure to synergy-maximizing integration that preserves and exploits complementary cross-modal signal.

bioinformatics↗

bioRxiv2026

Predicting evolutionary rate as a pretraining task improves genome language model representations

Genome language models (gLM) have the potential to further understanding of regulatory genomics without requiring labeled data. Most gLMs are pretrained using sequence reconstruction tasks inspired by natural language processing, but recent studies have shown that these gLMs often fail to capture biological signal. To overcome this, we introduce pretraining tasks that predict the rate of evolution. These tasks are designed so that they can be composed with sequence reconstruction, enabling a controlled comparison of predicting sequence only, evolutionary rate only, or both. To address gaps in existing evaluations, we developed a suite of biologically grounded benchmarks. Across these tasks, and for established variant effect prediction benchmarks, models pretrained on both sequence and evolutionary rate outperform those trained on sequence alone, and training on evolutionary rate can make the even the relatively small models in our work competitive with much larger existing gLMs for some tasks. These results establish evolution as a key training target for genome-scale models.

genomics↗

bioRxiv2025

Adaptive resampling for improved machine learning in imbalanced single-cell datasets

While machine learning models trained on single-cell transcriptomics data have shown great promise in providing biological insights, existing tools struggle to effectively model underrepresented and out-of-distribution cellular features or states. We present a generalizable Adaptive Resampling (AR) approach that addresses these limitations and enhances single-cell representation learning by resampling data based on its learned latent structure in an online, adaptive manner concurrent with model training. Experiments on gene expression reconstruction, cell type classification, and perturbation response prediction tasks demonstrate that the proposed AR training approach leads to significantly improved downstream performance across datasets and metrics. Additionally, it enhances the quality of learned cellular embeddings compared to standard training methods. Our results suggest that AR may serve as a valuable technique for improving representation learning and predictive performance in single-cell transcriptomic models.

bioinformatics↗

bioRxiv2025

Trainable subnetworks reveal insights into structure knowledge organization in protein language models

Protein language models (PLMs) pretrained via a masked language modeling objective have proven effective across a range of structure-related tasks, including high-resolution structure prediction. However, it remains unclear to what extent these models factorize protein structural categories among their learned parameters. In this work, we introduce trainable subnetworks, which mask out the PLM weights responsible for language modeling performance on a structural category of proteins. We systematically trained 39 PLM subnetworks targeting both sequence- and residue-level features at varying degrees of resolution using annotations defined by the CATH taxonomy and secondary structure elements. Using these PLM subnetworks, we assessed how structural factorization in PLMs influences downstream structure prediction. Our results show that PLMs are highly sensitive to sequence-level features and can predominantly disentangle extremely coarse or fine-grained information. Furthermore, we observe that structure prediction is highly responsive to factorized PLM representations and that small changes in language modeling performance can significantly impair PLM-based structure prediction capabilities. Our work presents a framework for studying feature entanglement within pretrained PLMs and can be leveraged to improve the alignment of learned PLM representations with known biological concepts.

bioinformatics↗

bioRxiv2025

Hierarchical cross-entropy loss improves atlas-scale single-cell annotation models

Accurately annotating cell types is essential for extracting biological insight from single-cell RNA-seq data. Although cell types are naturally organized into hierarchical ontologies, most computational models do not explicitly incorporate this structure into their training objectives. We introduce a hierarchical cross-entropy loss that aligns model objectives with biological structure. Applied to architectures ranging from linear models to transformers, this simple modification significantly improves out-of-distribution performance (12-15%) without added computational cost.

cell biology↗

bioRxiv2025

Characterizing Clinical Toxicity in Cancer Combination Therapies

AO_SCPLOWBSTRACTC_SCPLOWPredicting synergistic cancer drug combinations through computational methods offers a scalable approach to creating therapies that are more effective and less toxic. However, most algorithms focus solely on synergy without considering toxicity when selecting optimal drug combinations. In the absence of combinatorial toxicity assays, a few models use toxicity penalties to balance high synergy with lower toxicity. However, these penalties have not been explicitly validated against known drug-drug interactions. In this study, we examine whether synergy scores and toxicity metrics correlate with known adverse drug interactions. While some metrics show trends with toxicity levels, our results reveal significant limitations in using them as penalties. These findings highlight the challenges of incorporating toxicity into synergy prediction frameworks and suggest that advancing the field requires more comprehensive combination toxicity data.

bioinformatics↗

bioRxiv2025

Consequences of training data composition for deep learning models in single-cell biology

Foundation models for single-cell transcriptomics have the potential to augment (or replace) purpose-built tools for a variety of common analyses, especially when data are sparse. Recent work with large language models has shown that training data composition greatly shapes performance; however, to date, single-cell foundation models have ignored this aspect, opting instead to train on the largest possible corpus. We systematically investigate the consequences of training dataset composition on the behavior of deep learning models of single-cell transcriptomics, focusing on human hematopoiesis as a tractable model system and including cells from adult and developing tissues, disease states, and perturbation atlases. We find that (1) these models generalize poorly to unseen cell types, (2) adding malignant cells to a healthy cell training corpus does not necessarily improve modeling of unseen malignant cells, and (3) including an embryonic stem cell differentiation atlas during training improves performance on out-of-distribution tasks. Our results emphasize the importance of diverse training data and suggest strategies to optimize future single-cell foundation models.

genomics↗

bioRxiv2025

ML-MAGES: A machine learning framework for multivariate genetic association analyses with genes and effect size shrinkage

A fundamental goal of genetics is to identify which and how genetic variants are associated with a trait, often using the regression results from genome-wide association (GWA) studies. Important methodological challenges are accounting for inflation in GWA effect estimates as well as investigating more than one trait simultaneously. We leverage machine learning approaches for these two challenges, developing a computationally efficient method called ML-MAGES. First, we shrink the inflation in GWA effect sizes caused by non-independence among variants using neural networks. We then cluster variant associations among multiple traits via variational inference. We compare the performance of shrinkage via neural networks to regularized regression and fine-mapping, two approaches used for addressing inflated effects but dealing with variants in focal regions of different sizes. Our neural network shrinkage outperforms both methods in approximating the true effect sizes in simulated data. Our infinite mixture clustering approach offers a flexible, data-driven way to distinguish different types of associations--trait-specific, shared across traits, or spurious--among multiple traits based on their regularized effects. Clustering applied to our neural network shrinkage results also produces consistently higher precision and recall for distinguishing gene-level associations in simulations. We demonstrate the application of ML-MAGES on association analyses of two quantitative traits and two binary traits in the UK Biobank (genetic and phenotypic data from 500,000 residents of the UK). Our identified associated genes from single-trait enrichment tests overlap with those having known relevant biological processes to the traits. Besides trait-specific associations, ML-MAGES identifies several variants with shared multi-trait associations, suggesting putative shared genetic architecture.

genetics↗

bioRxiv2025

Sparse modeling of interactions enables fast detection of genome-wide epistasis in biobank-scale studies

The lack of computational methods capable of detecting epistasis in biobanks has led to uncertainty about the role of non-additive genetic effects on complex trait variation. The marginal epistasis framework is a powerful approach because it estimates the likelihood of a SNP being involved in any interaction, thereby reducing the multiple testing burden. Current implementations of this approach have failed to scale to large human studies. To address this, we present the sparse marginal epistasis (SME) test, which concentrates the scans for epistasis to regions of the genome that have known functional enrichment for a trait of interest. By leveraging the sparse nature of this modeling setup, we develop a novel statistical algorithm that allows SME to run 10 to 90 times faster than state-of-the-art epistatic mapping methods. In a study of blood traits measured in 349,411 individuals from the UK Biobank, we show that reducing searches of epistasis to variants in accessible chromatin regions facilitates the identification of genetic interactions associated with regulatory genomic elements.

genetics↗

bioRxiv2024

Rethinking cancer drug synergy prediction: a call for standardization in machine learning applications

Drug resistance poses a significant challenge to cancer treatment, often caused by intratumor heterogeneity. Combination therapies have been shown to be an effective strategy to prevent resistant cancer cells from escaping single-drug treatments. However, discovering new drug combinations through traditional molecular assays can be costly and time-consuming. In silico approaches can overcome this limitation by exploring many candidate combinations at scale. This study systematically evaluates the utility of various machine learning algorithms, input features, and drug synergy prediction tasks. Our findings indicate a pressing need for establishing a standardized framework to measure and develop algorithms capable of predicting synergy.

bioinformatics↗

bioRxiv2024

Evaluating the role of pre-training dataset size and diversity on single-cell foundation model performance

The success of transformer-based foundation models on natural language and images has motivated their use in single-cell biology. Single-cell foundation models have been trained on increasingly larger transcriptomic datasets, scaling from initial studies with 1 million cells to newer atlases with over 100 million cells. This study investigates the role of pre-training dataset size and diversity on the performance of single-cell foundation models on both zero-shot and fine-tuned tasks. Using a large corpus of 22.2 million cells, we pre-train a total of 400 models, which we evaluate by conducting 6,400 experiments. Our results show that current methods tend to plateau in performance with pre-training datasets that are only a fraction of the size of current training corpora.

genomics↗

bioRxiv2024

Sex differences in the cardiopulmonary and neuromuscular response to high-intensity interval exercise

Sex differences exist in the integrative response to exercise, however, these are typically researched during constant-load exercise. Interval exercise involves high-intensity efforts interspersed with recovery periods to repeatedly stress physiological systems, and it is currently unknown whether the response to this form of exercise differs between sexes. Ten males and ten females (age: 25{+/-}3 years) completed two experimental visits. First, an incremental treadmill exercise test was performed to obtain submaximal (lactate threshold) and maximal ([Formula] O2peak) data. Thereafter, visit two involved 4 x 3-min running intervals at 90% of the final incremental test velocity (v[Formula] O2peak), with 90 secs rest between intervals. Before exercise and after each interval, maximal voluntary contraction (MVC), quadriceps potentiated twitch (Qtw.pot), and voluntary activation (VA) were recorded. The rates of oxygen uptake ([Formula] O2), carbon dioxide production ([Formula] CO2) and ventilation ([Formula] E) were continuously recorded throughout. There was no sex difference in relative [Formula] O2peak (males: 47.2{+/-}6.0 vs. females: 44.4{+/-}5.8 ml.kg- 1.min-1, p=0.292). When expressed relative to peak values, there were no sex differences in the [Formula] O2 or [Formula] CO2 response to the interval task (p[≥]0.781). Females had greater [Formula] E, [Formula] E/[Formula] O2, and [Formula] E/[Formula] CO2 values during the first two intervals (p[≤]0.046). There were no sex differences in the reductions in MVC, Qtw.pot, and VA during the interval task (p[≥]0.150), however females had lesser reductions in Qtw.pot values post-exercise (-24{+/-}9 vs. -15{+/-}8%, p=0.044). Sex differences exist in the physiological response to interval exercise. Compared to males, females experienced greater hyperpnoea during the initial stages, and had lesser decreases in contractile function post-exercise.

physiology↗

bioRxiv2024

Mutation and cell state compatibility is required and targetable in Ph+ acute lymphoblastic leukemia minimal residual disease

Efforts to cure BCR::ABL1 B cell acute lymphoblastic leukemia (Ph+ ALL) solely through inhibition of ABL1 kinase activity have thus far been insufficient despite the availability of tyrosine kinase inhibitors (TKIs) with broad activity against resistance mutants. The mechanisms that drive persistence within minimal residual disease (MRD) remain poorly understood and therefore untargeted. Utilizing 13 patient-derived xenograft (PDX) models and clinical trial specimens of Ph+ ALL, we examined how genetic and transcriptional features co-evolve to drive progression during prolonged TKI response. Our work reveals a landscape of cooperative mutational and transcriptional escape mechanisms that differ from those causing resistance to first generation TKIs. By analyzing MRD during remission, we show that the same resistance mutation can either increase or decrease cellular fitness depending on transcriptional state. We further demonstrate that directly targeting transcriptional state-associated vulnerabilities at MRD can overcome BCR::ABL1 independence, suggesting a new paradigm for rationally eradicating MRD prior to relapse. Finally, we illustrate how cell mass measurements of leukemia cells can be used to rapidly monitor dominant transcriptional features of Ph+ ALL to help rationally guide therapeutic selection from low-input samples. HIGHLIGHTSO_LIRelapse after remission on TKI can harbor mutations in ABL1, RAS, or neither C_LIO_LIMutations and development-like cell state dictate fitness in residual disease C_LIO_LICo-targeting cell state and ABL1 markedly reduces MRD C_LIO_LIBiophysical measurements provide an integrative, rapid measurement of cell state C_LI

cancer biology↗

bioRxiv2024

Incorporation of Epstein-Barr viral variation implicates significance of LMP1 in survival prediction and prognostic subgrouping in Burkitt lymphoma

While Epstein-Barr virus (EBV) plays a role in Burkitt lymphoma (BL) tumorigenesis, it is unclear if EBV genetic variation impacts clinical outcomes. From 130 publicly available whole-genome tumor sequences of EBV-positive BL patients, we used least absolute shrinkage and selection operator (LASSO) regression and Bayesian variable selection models within a Cox proportional hazards framework to select the top EBV variants, putative driver genes, and clinical features associated with patient survival time. These features were incorporated into survival prediction and prognostic subgrouping models. Our model yielded 22 EBV variants including seven in LMP1 as most associated with patient survival time. Using the top EBV variants, driver genes, and clinical features, we defined three prognostic subgroups that demonstrated differential survival rates, laying the foundation for incorporating EBV variants such as those in LMP1 as predictive biomarker candidates in future studies.

genomics↗