Shankar Subramaniam
Research Interests
Make your next move.
Prepare your application
Review drafts before you send them.
Explore related searches
Articles (5)
Heterogeneity of breast cancer poses several challenges for detection and treatment. With next-generation sequencing, we can now map the transcriptional profile of each patient’s breast tissue, which has the potential for identifying and characterizing cancer subtypes. However, the large dimensionality of this transcriptomic data and the heterogeneity between the molecular profiles of breast cancers poses a barrier to identifying minimal markers and mechanistic consequences. In this study, we develop an autoencoder to identify a reduced set of gene markers that characterize the four major breast cancer subtypes with the accuracy of 82.38%. The reduced feature space created by our model captures the functional characteristics of each breast cancer subtype highlighting mechanisms that are unique to each subtype as well as those that are shared. Our high prediction accuracy shows that our markers can be valuable for breast cancer subtype detection and have the potential to provide insights into mechanisms associated with each subtype.
Year:
2025
Recent applications of foundation models in biology have focused on pretraining using large-scale single-cell datasets comprising millions of cells, across diverse patho-physiological states. These models are then fine-tuned for downstream tasks such as cell-type classification. In this study, we evaluated the performance of three widely-used foundation models in biology—scGPT, SCMAMBA-2, and Geneformer—and a statistical baseline (Seurat v5) on cell-type classification under Gaussian noise perturbation. We used two curated datasets, referred to as Myeloid (13k cells) and hPancreas (15k cells). Surprisingly, we found that the baseline performance of the foundation models was inferior to that of the statistical model, even without any added perturbation. Although we note that model size can affect performance, Geneformer’s accuracy outperformed scGPT and SCMAMBA-2 by 5% on average across all datasets despite having 40% fewer trainable parameters. Nonetheless, the statistical baseline still outperformed Geneformer by 9% in accuracy. Based on these findings, we hypothesized that the conventional training paradigm used by foundation models for single-cell tasks consistently underperforms statistical models due to a lack of essential biological context. To investigate this, we evaluated whether performance degradation stems from early data embedding steps—such as binning or gene normalization during tokenization, and sampling bias. First, to better understand tokenization-related artifacts, we introduced controlled Gaussian noise to gene expression values before tokenization, amplifying downstream distortions introduced by the tokenization process (all models were trained for identical step durations with identical hyperparameters). On the Myeloid dataset, following the introduction of Gaussian noise perturbation to 20% of cells, both scGPT and SCMAMBA-2 saw an 11% decrease in accuracy while Geneformer saw an 8% decrease in accuracy. This difference in performance may be due to the difference in encoding methods used by both models. scGPT and SCMAMBA-2 use a bin-based tokenization strategy, in contrast to Geneformer’s rank-value encoding, which normalizes gene expression values using a predetermined encoding constant. Although binning captures general trends in count data, it fails to preserve relative expression at the gene level, resulting in significant information loss. Applying the three models to the Myeloid dataset also revealed that scGPT’s prediction distribution is biased toward overrepresented cell types in the training data, while underrepresenting rarer classes. CD14 cells are overpredicted by 16% (among most abundant cell types) by scGPT. Geneformer, however, maintains a more stable prediction distribution with a 6% (CD14) increase and outperforms scGPT and SCMAMBA-2 by 26% in macro F1 score (unweighted metric). Based on our findings we assert that the gap in contextual encoding in bin-based tokenization is what contributes to the less-nuanced learning. Recent research that integrated cellular-ontology during training showed improved performance to both scGPT and Geneformer. Our results underscore a fundamental issue, that foundation models lack critical biological context that would allow for them to make the nuanced inferences required for complex biological analyses. The compression of single-cell data from raw counts to embedding vectors can span several orders of magnitude, and lead to significant loss of information. As a result, methods must adapt to prioritize contextual integration during tokenization to ensure sufficient information for the model.
Year:
2025
Background Triple-negative breast cancer is a prevalent breast cancer subtype with the lowest 5-year survival. Several factors contribute to its treatment response, but the inherent molecular and cellular tumor heterogeneity are increasingly acknowledged as crucial determinants. Methods Spatial transcriptomic profiling was performed on FFPE tissues from a retrospective, treatment-naive group of women with differential prognoses (17 with >15 years survival-good prognosis (GPx) and 15 with <3 years survival-poor prognosis (PPx)) using GeoMX ® Digital Spatial Profiler. Regions of interest were segmented on pan-cytokeratin and analyzed for tumor and stromal components, probed using GeoMx human whole transcriptome atlas (WTA) panel. Data quality control, normalization, and differential analysis was performed in R using GeomxTools and linear mixed models. Additional analyses including cell-type deconvolution, spatial entropy, functional enrichment, TF-target / ligand-receptor analysis and convolution neural networks were employed to identify significant gene signatures contributing to differential prognosis. Results Here we report on the spatial and molecular heterogeneity underlying differential prognosis. We observe that the state of the epithelia and its microenvironment (TME) are transcriptionally distinct between the two groups. Invasive epithelia in GPx show a significant increase in immune transcripts with the TME exhibiting increased immune cell presence (via IF), while in PPx they are more metabolically and translationally active, with the TME being more mesenchymal/fibrotic. Specifically, pre-cancerous epithelia in PPx display a prescience of aggressiveness as evidenced by increased EMT-signaling. We identify distinct epithelial gene signatures for PPx and GPx, that can, with high accuracy, classify samples at the time of diagnosis and likely inform therapy. Conclusions To the best of our knowledge, this is the first study to leverage spatial transcriptomics for an in-depth delineation of the cellular and molecular underpinnings of differential prognosis in TNBC. Our study highlights the potential of spatial transcriptomics to not only uncover the molecular drivers of differential prognosis in TNBC but also to pave the way for precision diagnostics and tailored therapeutic strategies, transforming the clinical landscape for this aggressive breast cancer subtype.
Year:
2025
Many biomedical research projects produce large-scale datasets that may serve as resources for the research community for hypothesis generation, facilitating diverse use cases. Towards the goal of developing infrastructure to support the findability, accessibility, interoperability, and reusability (FAIR) of biomedical digital objects and maximally extracting knowledge from data, complex queries that span across data and tools from multiple resources are currently not easily possible. By utilizing existing FAIR application programming interfaces (APIs) that serve knowledge from many repositories and bioinformatics tools, different types of complex queries and workflows can be created by using these APIs together. The Playbook Workflow Builder (PWB) is a web-based platform that facilitates interactive construction of workflows by enabling users to utilize an ever-growing network of input datasets, semantically annotated API endpoints, and data visualization tools contributed by an ecosystem. Via a user-friendly web-based user interface (UI), workflows can be constructed from contributed building-blocks without technical expertise. The output of each step of the workflows are provided in reports containing textual descriptions, as well as interactive and downloadable figures and tables. To demonstrate the ability of the PWB to generate meaningful hypotheses that draw knowledge from across multiple resources, we present several use cases. For example, one of these use cases sieves novel targets for individual cancer patients using data from the GTEx, LINCS, Metabolomics, GlyGen, and the ExRNA Communication Consortium (ERCC) Common Fund (CF) Data Coordination Centers (DCCs). The workflows created with the PWB can be published and repurposed to tackle similar use cases using different inputs. The PWB platform is available from: https://playbook-workflow-builder.cloud/ .
Year:
2024
Background Biomedical research often involves contextual integration of multimodal and multiomic data in search of mechanisms for improved diagnosis, treatment, and monitoring. Researchers need to access information from diverse sources, comprising data in various and sometimes incongruent formats. The downstream processing of the data to decipher mechanisms by reconstructing networks and developing quantitative models warrants considerable effort. Results MetGENE is a knowledge-based, gene-centric data aggregator that hierarchically retrieves information about the gene(s), their related pathway(s), reaction(s), metabolite(s), and metabolomic studies from standard data repositories under one dashboard to enable ease of access through centralization of relevant information. We note that MetGENE focuses only on those genes that encode for proteins directly associated with metabolites. All other gene–metabolite associations are beyond the current scope of MetGENE. Further, the information can be contextualized by filtering by species, anatomy (tissue), and condition (disease or phenotype). Conclusions MetGENE is an open-source tool that aggregates metabolite information for a given gene(s) and presents them in different computable formats (e.g., JSON) for further integration with other omics studies. MetGENE is available at https://bdcw.org/MetGENE/index.php.
Year:
2022
Collaborators (8)
Raja Mazumder
George Washington University
Adam Resnick
Director
University of Pennsylvania
Jonathon Keeney
Assistant Research Professor
George Washington University
Aleksandar Mihajlovic
Head of Operations
APIS Assay Technologies Ltd
Deanne Taylor
Research Associate Professor
University of Pennsylvania
Alexander Lachmann
Assistant Professor
Icahn School of Medicine at Mount Sinai
Aleksandar Milosavljevic
Baylor College of Medicine
Avi Ma'ayan
Professor
Icahn School of Medicine at Mount Sinai
Make your next move.
Prepare your application
Review drafts before you send them.