Feature attribution, the ability to localize regions of the input data that are relevant for classification, is an important capability for ML models in scientific and biomedical domains. Current methods for feature attribution, which rely on "explaining" the predictions of end-to-end classifiers, suffer from imprecise feature localization and are inadequate for use with small sample sizes and high-dimensional datasets due to computational challenges. We introduce prospector heads, an efficient and interpretable alternative to explanation-based attribution methods that can be applied to any encoder and any data modality. Prospector heads generalize across modalities through experiments on sequences (text), images (pathology), and graphs (protein structures), outperforming baseline attribution methods by up to 26.3 points in mean localization AUPRC. We also demonstrate how prospector heads enable improved interpretation and discovery of class-specific patterns in input data. Through their high performance, flexibility, and generalizability, prospectors provide a framework for improving trust and transparency for ML models in complex domains.

BACKGROUND: While technological advances have made it possible to profile the immune system at high resolution, translating high-throughput data into knowledge of immune mechanisms has been challenged by the complexity of the interactions underlying immune processes. Tools to explore the immune network are critical for better understanding the multi-layered processes that underlie immune function and dysfunction, but require a standardized network map of immune interactions. To facilitate this we have developed ImmunoGlobe, a manually curated intercellular immune interaction network extracted from Janeway's Immunobiology textbook. RESULTS: ImmunoGlobe is the first graphical representation of the immune interactome, and is comprised of 253 immune system components and 1112 unique immune interactions with detailed functional and characteristic annotations. Analysis of this network shows that it recapitulates known features of the human immune system and can be used uncover novel multi-step immune pathways, examine species-specific differences in immune processes, and predict the response of immune cells to stimuli. ImmunoGlobe is publicly available through a user-friendly interface at www.immunoglobe.org and can be downloaded as a computable graph and network table. CONCLUSION: While the fields of proteomics and genomics have long benefited from network analysis tools, no such tool yet exists for immunology. ImmunoGlobe provides a ground truth immune interaction network upon which such tools can be built. These tools will allow us to predict the outcome of complex immune interactions, providing mechanistic insight that allows us to precisely modulate immune responses in health and disease.

Recent advances in our understanding of cancer progression have highlighted the roles played by molecular heterogeneity and by the tumor microenvironment in driving drug resistance and metastasis. The coupling of single-cell measurement technologies with algorithms, such as t-sne and SPADE, have enabled deep investigation of tumor heterogeneity. However, such techniques only capture molecular heterogeneity and do not enable the quantification nor visualization of intercellular interactions. They additionally do not allow the visualization of ecological niches that are critical to understanding tumor behavior. Novel computational tools to quantify and visualize spatial patterns in the tumor microenvironment are critically needed. Here, we take a tumor ecology perspective to examine how predation, mutualism, commensalism, and parasitism may impact tumor development and spatial patterning. We additionally quantify local spatial heterogeneity and the emergent global spatial behavior of the models using geostatistics. By visualizing emergent spatial patterns we demonstrate the potential utility of a geostatistical analysis in differentiating amongst cell-cell interactions in the tumor microenvironment. These studies introduce both an ecological framework for characterizing intercellular interactions in cancer and a novel way of quantifying and visualizing spatial patterns in cancer.

Scientific data is continuously generated throughout the world. However, analyses of these data are typically performed exactly once and on a small fragment of recently generated data. Ideally, data analysis would be a continuous process that uses all the data available at the time, and would be automatically re-run and updated when new data appears. We present a framework for automated discovery from data repositories that tests user-provided hypotheses using expert-grade data analysis strategies, and reassesses hypotheses when more data becomes available. Novel contributions of this approach include a framework to trigger new analyses appropriate for the available data through lines of inquiry that support progressive hypothesis evolution, and a representation of hypothesis revisions with provenance records that can be used to inspect the results. We implemented our approach in the DISK framework, and evaluated it using two scenarios from cancer multi-omics: 1) data for new patients becomes available over time, 2) new types of data for the same patients are released. We show that in all scenarios DISK updates the confidence on the original hypotheses as it automatically analyzes new data.

Abstract Cardinal is an R package for statistical analysis of mass spectrometry-based imaging (MSI) experiments of biological samples such as tissues. Cardinal supports both Matrix-Assisted Laser Desorption/Ionization (MALDI) and Desorption Electrospray Ionization-based MSI workflows, and experiments with multiple tissues and complex designs. The main analytical functionalities include (1) image segmentation, which partitions a tissue into regions of homogeneous chemical composition, selects the number of segments and the subset of informative ions, and characterizes the associated uncertainty and (2) image classification, which assigns locations on the tissue to pre-defined classes, selects the subset of informative ions, and estimates the resulting classification error by (cross-) validation. The statistical methods are based on mixture modeling and regularization. Contact: [email protected] Availability and implementation: The code, the documentation, and examples are available open-source at www.cardinalmsi.org under the Artistic-2.0 license. The package is available at www.bioconductor.org.

SUMMARY: The ProteoWizard software project provides a modular and extensible set of open-source, cross-platform tools and libraries. The tools perform proteomics data analyses; the libraries enable rapid tool creation by providing a robust, pluggable development framework that simplifies and unifies data file access, and performs standard proteomics and LCMS dataset computations. The library contains readers and writers of the mzML data format, which has been written using modern C++ techniques and design principles and supports a variety of platforms with native compilers. The software has been specifically released under the Apache v2 license to ensure it can be used in both academic and commercial projects. In addition to the library, we also introduce a rapidly growing set of companion tools whose implementation helps to illustrate the simplicity of developing applications on top of the ProteoWizard library. AVAILABILITY: Cross-platform software that compiles using native compilers (i.e. GCC on Linux, MSVC on Windows and XCode on OSX) is available for download free of charge, at http://proteowizard.sourceforge.net. This website also provides code examples, and documentation. It is our hope the ProteoWizard project will become a standard platform for proteomics development; consequently, code use, contribution and further development are strongly encouraged.