Current challenges in developing foundational models for volumetric imaging data, such as magnetic resonance imaging (MRI), stem from the computational complexity of state-of-the-art architectures in high dimensions and curating sufficiently large datasets of volumes. To address these challenges, we introduce Raptor (Random Planar Tensor Reduction), a train-free method for generating semantically rich embeddings for volumetric data. Raptor leverages a frozen 2D foundation model, pretrained on natural images, to extract visual tokens from individual cross-sections of medical volumes. These tokens are then spatially compressed using random projections, significantly reducing computational complexity while retaining rich semantic information. Extensive experiments on 10 diverse medical volume tasks verify the superior performance of Raptor over state-of-the-art methods, including those pretrained exclusively on medical volumes (+3% SuPreM, +6% MISFM, +10% Merlin, +13% VoCo, and +14% SLIViT), while entirely bypassing the need for costly training. Our results highlight Raptor’s effectiveness and versatility as a foundation for advancing deep learning-based methods for medical volumes (code: github.com/sriramlab/raptor).

We present CACTI, a masked autoencoding approach for imputing tabular data that leverages the structure in missingness patterns and contextual information. Our approach employs a novel median truncated copy masking training strategy that encourages the model to learn from empirical patterns of missingness while incorporating semantic relationships between features — captured by column names and text descriptions — to better represent feature dependence. These dual sources of inductive bias enable CACTIto outperform state-of-the-art methods — an average $R^2$ gain of 7.8\% over the next best method (13.4%, 6.1%, and 5.3% under missing not at random, at random and completely at random, respectively) — across a diverse range of datasets and missingness conditions. Our results highlight the value of leveraging dataset-specific contextual information and missingness patterns to enhance imputation performance.

MOTIVATION: Since the first human genome was sequenced in 2001, there has been a rapid growth in the number of bioinformatic methods to process and analyze next-generation sequencing (NGS) data for research and clinical studies that aim to identify genetic variants influencing diseases and traits. To achieve this goal, one first needs to call genetic variants from NGS data, which requires multiple computationally intensive analysis steps. Unfortunately, there is a lack of an open-source pipeline that can perform all these steps on NGS data in a manner, which is fully automated, efficient, rapid, scalable, modular, user-friendly and fault tolerant. To address this, we introduce xGAP, an extensible Genome Analysis Pipeline, which implements modified GATK best practice to analyze DNA-seq data with the aforementioned functionalities. RESULTS: xGAP implements massive parallelization of the modified GATK best practice pipeline by splitting a genome into many smaller regions with efficient load-balancing to achieve high scalability. It can process 30× coverage whole-genome sequencing (WGS) data in ∼90 min. In terms of accuracy of discovered variants, xGAP achieves average F1 scores of 99.37% for single nucleotide variants and 99.20% for insertion/deletions across seven benchmark WGS datasets. We achieve highly consistent results across multiple on-premises (SGE & SLURM) high-performance clusters. Compared to the Churchill pipeline, with similar parallelization, xGAP is 20% faster when analyzing 50× coverage WGS on Amazon Web Service. Finally, xGAP is user-friendly and fault tolerant where it can automatically re-initiate failed processes to minimize required user intervention. AVAILABILITY AND IMPLEMENTATION: xGAP is available at https://github.com/Adigorla/xgap. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.