ML infrastructure and data systems for precision oncology

Discovering and accessing relevant biomedical data across heterogeneous repositories remains a significant bottleneck in research workflows. Traditional schema-based retrieval systems require predefined metadata structures, limiting their applicability to diverse data sources. We present DeepGEOSearch, an LLM-powered retrieval system that enables semantic search and discovery of biomedical datasets through natural language queries without requiring explicit schema definitions. Our approach leverages large language models for flexible data interpretation and ranking, allowing researchers to discover datasets based on experimental intent rather than rigid schema constraints. We demonstrate the system's effectiveness on large-scale biomedical data repositories, showing improved discoverability and usability compared to traditional keyword-based approaches.

Recent research has shown the impressive capability of large language models like GPT-4 in various downstream tasks in single-cell data analysis. Among these tasks, cell type annotation remains particularly challenging, with researchers exploring various methods to improve accuracy and efficiency. While recent studies on GPT-like models have demonstrated annotation performance comparable to manual annotations, a significant gap remains in understanding their limitations and generalizability. In this work, we compare and evaluate the annotation performance of the GPT-4 model against traditional methods on nine randomly selected public single-cell RNA seq datasets from cellxgene, covering diverse tissue types. Our evaluation highlights the complexity of annotating cell types in single-cell data, revealing key differences between automated and manual approaches. We found specific cases where GPT-4 underperforms, demonstrating its limitations in certain contexts. We further introduce an automated approach to incorporate literature search using a RAG approach which enhances and outperforms GPT-4 cell type annotation when compared to traditional methods. We also introduce metrics based on taxonomic distance in the ontology tree to evaluate the granularity of the cell type annotations. To support future research, we also release an open-source Python package that enables fully automated cell-type annotation of single-cell data using GPT-4 alongside other methods. The pipeline can take paper as an input and do cell type annotations on its own.

Recently, fine-tuned transformer-based models (eg, PubMedBERT, BioBERT) have shown the state-of-the-art performance of a number of BioNLP tasks, such as Named Entity Recognition (NER). However, transformer-based models are complex and have millions of parameters, and, consequently, are relatively slow during inference. In this paper, we address the time complexity limitations of the BioNLP transformer models. In particular, we propose a Multi-Task Learning based framework for jointly learning three different biomedical NER tasks. Our experiments show a reduction in inference time by a factor of three without any reduction in prediction accuracy.

This work deploys linguistically motivated features to classify paragraph-level text into fiction and non-fiction genre using a logistic regression model and infers lexical and syntactic properties that distinguish the two genres. Previous works have focused on classifying document-level text into fiction and non-fiction genres, while in this work, we deal with shorter texts which are closer to real-world applications like sentiment analysis of tweets. Going beyond simple POS tag ratios proposed in Qureshi et al.(2019) for document-level classification, we extracted multiple linguistically motivated features belonging to four categories: Lexical features, POS ratio features, Syntactic features and Raw features. For the task of short-text classification, a model containing 28 best-features (selected via Recursive feature elimination with cross-validation; RFECV) confers an accuracy jump of 15.56 % over a baseline model consisting of 2 POS-ratio features found effective in previous work (cited above). The efficacy of the above model containing a linguistically motivated feature set also transfers over to another dataset viz, Baby BNC corpus. We also compared the classification accuracy of the logistic regression model with two deep-learning models. A 1D CNN model gives an increase of 2% accuracy over the logistic Regression classifier on both corpora. And the BERT-base-uncased model gives the best classification accuracy of 97% on Brown corpus and 98% on Baby BNC corpus. Although both the deep learning models give better results in terms of classification accuracy, the problem of interpreting these models remains unsolved. In contrast, regression model coefficients revealed that fiction texts tend to have more character-level diversity and have lower lexical density (quantified using content-function word ratios) compared to non-fiction texts. Moreover, subtle differences in word order exist between the two genres, i.e., in fiction texts Verbs precede Adverbs (inter-alia).