Immune-related processes are important in underpinning the properties of clinical traits such as prognosis and drug response in cancer. The possibility to extract knowledge learned by artificial neural networks (ANNs) from omics data to explain cancer clinical traits is a very attractive subject for novel discovery. Recent studies using a version of ANNs called autoencoders revealed their capability to store biologically meaningful information indicating that autoencoders can be utilized as knowledge discovery platforms aside from their initial assigned use for dimensionality reduction. Here, we devise an innovative weight engineering approach and ANN platform called artificial neural network encoder (ANNE) using an autoencoder and apply it to a breast cancer dataset to extract knowledge learned by the autoencoder model that explains clinical traits. Intriguingly, the extracted biological knowledge in the form of gene–gene associations from ANNE shows immune-related components such as chemokines, carbonic anhydrase, and iron metabolism that modulate immune-related processes and the tumor microenvironment play important roles in underpinning breast cancer clinical traits. Our work shows that biological “knowledge” learned by an ANN model is indeed encoded as weights throughout its neuronal connections, and it is possible to extract learned knowledge via a novel weight engineering approach to uncover important biological insights.

Anticipating and understanding cancers’ need for specific gene activities is key for novel therapeutic development. Here we utilized DepMap, a cancer gene dependency screen, to demonstrate that machine learning combined with network biology can produce robust algorithms that both predict what genes a cancer is dependent on and what network features coordinate such gene dependencies. Using network topology and biological annotations, we constructed four groups of novel engineered machine learning features that produced high accuracies when predicting binary gene dependencies. We found that in all examined cancer types, F1 scores were greater than 0.90, and model accuracy remained robust under multiple hyperparameter tests. We then deconstructed these models to identify tumor type-specific coordinators of gene dependency and identified that in certain cancers, such as thyroid and kidney, tumors’ dependencies are highly predicted by gene connectivity. In contrast, other histologies relied on pathway-based features such as lung, where gene dependencies were highly predictive by associations with cell death pathway genes. In sum, we show that biologically informed network features can be a valuable and robust addition to predictive pharmacology models while simultaneously providing mechanistic insights.

Spatially resolved sequencing technologies help us dissect how cells are organized in space. Several available computational approaches focus on the identification of spatially variable genes (SVGs), genes whose expression patterns vary in space. The detection of SVGs is analogous to the identification of differentially expressed genes and permits us to understand how genes and associated molecular processes are spatially distributed within cellular niches. However, the expression activities of SVGs fail to encode all information inherent in the spatial distribution of cells. Here, we devised a deep learning model, Spatially Informed Artificial Intelligence (SPIN-AI), to identify spatially predictive genes (SPGs), whose expression can predict how cells are organized in space. We used SPIN-AI on spatial transcriptomic data from squamous cell carcinoma (SCC) as a proof of concept. Our results demonstrate that SPGs not only recapitulate the biology of SCC but also identify genes distinct from SVGs. Moreover, we found a substantial number of ribosomal genes that were SPGs but not SVGs. Since SPGs possess the capability to predict spatial cellular organization, we reason that SPGs capture more biologically relevant information for a given cellular niche than SVGs. Thus, SPIN-AI has broad applications for detecting SPGs and uncovering which biological processes play important roles in governing cellular organization.

High grade glioma (HGG) represents a group of devastating diseases with dismal prognosis. Surgical resection of the contrast enhancing (CE) region of HGG remains the mainstay of treatment, but recurrence inevitably arises from the unresected non-contrast enhancing (NE) region, surgically inaccessible due to cancer cell invasion into healthy brain tissue. Due to its critical role in recurrence, understanding of the NE region is central to the improvement of clinical outcomes. We reveal the biological characteristics of this region through image localized multi-regional sampling. We linked microenvironmental characteristics measured by multi-parametric MRI to genomic mutations and transcriptional phenotypes using mixed effect modeling which allowed us to control for individualized patient effects. We first confirmed that T2 is a significant indicator of IDH mutation status in the NE region, being the first description of such a relationship in a HGG cohort. We found the combination of EGFR amplification and CDKN2A homozygous loss was associated with a significantly lower mean diffusivity (MD) compared to double wild type tumors in the NE region, indicating the presence of greater cellular packing and proliferation in EGFR amplification/CDKN2A loss regions. Finally, using single cell pathway based tumor classifications, we showed that nK2, a DSC-MRI metric representing cell size heterogeneity, correlated positively with neuronal signature and negatively with glycolytic/plurimetabolic signature within the NE tumor, indicating that glycolytic/plurimetobolic tumors possessed a high amount of cell size heterogeneity compared to neuronal samples. This hypothesis was supported using digital reference object (DRO) modeling which confirmed that cell size and heterogeneity drove the differential nK2 signal between neuronal and glycolytic/plurimetabolic samples. We identified immune cell infiltrate as one possible mechanism of increased cell size heterogeneity using transcriptomic signature analysis which found more immune cell signatures within glycolytic/plurimetobolic tumors compared to neuronal. Collectively this study demonstrates the central role of multi-parametric MRI as a non-invasive measure of tumor biology and a tool for understanding the clinically critical NE region which can then inform new therapies targeting this region of HGG recurrence. Citation Format: Matthew Flick, Taylor Weiskittel, Kevin Meng-Lin, Fulvio D'Angelo, Francesca Caruso, Shannon Ensign, Mylan Blomquist, Luija Wang, Christopher Sereduk, Gustavo De Leon, Ashley Nespodzany, Javier Urcuyo, Ashlynn Gonzalez, Lee Curtin, Kyle Singleton, Aliya Anil, Natenael Simmineh, Erika Lewis, Teresa Noviello, Reyna Patel, Panwen Wang, Junwen Wang, Jennifer Eschbacher, Andrea Hawkins-Daarud, Pamela Jackson, Kris Smith, Peter Nakaji, Bernard Bendok, Richard Zimmerman, Chandan Krishna, Devi Patra, Naresh Patel, Mark Lyons, Matthew Neal, Kliment Donev, Maciej Mrugala, Alyx Porter, Scott Beeman, Yuxiang Zhou, Leslie Baxter, Christopher Plaisier, Jing Li, Hu Li, Anna Lasorella, Chad Quarles, Kristin Swanson, Michele Ceccarelli, Antonio Iavarone, Nhan Tran, Leland Hu. Multi-parametric MRI maps regional heterogeneity of high grade glioma phenotypes. [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2023; Part 1 (Regular and Invited Abstracts); 2023 Apr 14-19; Orlando, FL. Philadelphia (PA): AACR; Cancer Res 2023;83(7_Suppl):Abstract nr 5621.