Soumyadeep Roy

Large Language Models (LLMs) recently achieved great success in medical text summarization tasks by simply using in-context learning. However, these studies do not perform fine-grained evaluations under difficult settings where LLMs might fail. They typically report performance scores over the entire dataset. Through our benchmarking study, we show that LLMs show a significant performance drop for data points with high concentration of out-of-vocabulary (OOV) words or with high novelty (23.40% and 28.73% drop for Llama-2 and Llama-3.1 respectively). Vocabulary adaptation is an intuitive solution to this vocabulary mismatch issue where the LLM vocabulary gets updated with certain expert domain (here, medical) words. An interesting finding from our study is that Llama-3.1, even with a vocabulary size of around 128K tokens, still faces the over-fragmentation issue with medical words. We show vocabulary adaptation helps improve the LLM summarization performance. Through extensive experimentation of multiple vocabulary adaptation strategies, two continual pretraining strategies, and three benchmark medical summarization datasets, we gain valuable insights into the role of vocabulary adaptation strategies for customizing LLMs to the medical domain. We also performed a human evaluation study with medical experts where they found that vocabulary adaptation results in more relevant and faithful summaries.

This study presents a comprehensive benchmarking of three state-of-the-art single-cell foundation models scGPT, Geneformer, and scFoundation, on cell-type classification tasks. We evaluate the models on three datasets: myeloid, human pancreas, and multiple sclerosis, examining both standard fine-tuning and few-shot learning scenarios. Our work reveals that scFoundation consistently achieves the best performance while Geneformer performs poorly, yielding results sometimes even worse than those of the baseline models. Additionally, we demonstrate that a good foundation model can generalize well even when fine-tuned with out-of-distribution data, a capability that the baseline models lack. Our work highlights the potential of foundation models for addressing challenging biomedical questions, particularly in contexts where models are trained on one population but deployed on another.

AI is emerging as an efficient companion in medicine. While AI holds promise for reducing the cognitive load of researchers and practitioners, its adoption is often hindered by a lack of trust in new AI advancements. We present sophisticated techniques for developing trustworthy artificial intelligence (AI) models in medicine, bridging breakthroughs in AI research with practical healthcare applications. We will discuss in-depth the four stages (Design, Development, Implementation, and Evaluation) involved in the process of building trustworthy AI models customized for the medical domain. We present various techniques for incorporating important Trustworthy AI principles like data privacy, robustness, explainability, interpretability, medical experts-in-the-loop, and risk assessment while developing AI models for medicine. In contrast to prior tutorials, we make the following two key contributions: (i) While explaining the ’Implementation’ stage, we cover various real-world healthcare applications developed as part of research projects in academia in collaboration with medical schools in India and Germany. (ii) By including a health informatics professional as one of the tutorial organizers, we provide a fresh and much-needed perspective on the research challenges and mitigation strategies in building AI models for medicine.

In this work, we show a fundamental limitation in vocabulary adaptation approaches that use Byte-Pair Encoding (BPE) tokenization scheme for fine-tuning pretrained language models (PLMs) to expert domains. Current approaches trivially append the target domain-specific vocabulary at the end of the PLM vocabulary. This approach leads to a lower priority score and causes sub-optimal tokenization in BPE that iteratively uses merge rules to tokenize a given text. To mitigate this issue, we propose AdaptBPE where the BPE tokenization initialization phase is modified to first perform the longest string matching on the added (target) vocabulary before tokenizing at the character level. We perform an extensive evaluation of AdaptBPE versus the standard BPE over various classification and summarization tasks; AdaptBPE improves by 3.57% (in terms of accuracy) and 1.87% (in terms of Rouge-L), respectively. AdaptBPE for MEDVOC works particularly well when reference summaries have high OOV concentration or are longer in length. We also conduct a human evaluation, revealing that AdaptBPE generates more relevant and more faithful summaries as compared to MEDVOC.

Gene transformer models such as Nucleotide Transformer, DNABert, and LOGO are trained to learn optimal gene sequence representations, by using the Masked Language Modeling (MLM) training objective over the complete Human Reference Genome. However, the typical tokenization methods employ a basic sliding window of tokens, such as k-mers, that fail to utilize any gene-centric semantics. This could result in the (trivial) masking of sequences that are easily predictable, leading to inefficient MLM training. Time-variant training strategies are known to improve pretraining efficiency in both language and vision tasks. In this work, we focus on using curriculum masking where we systematically increase the difficulty of masked token prediction task, by using a Pointwise Mutual Information-based difficulty criterion, as gene sequences lack well-defined semantic units similar to words or sentences of NLP domain. Our proposed Curriculum Masking-based Gene Masking Strategy (CM-GEMS) demonstrates superior representation learning capabilities compared to baseline masking approaches, when evaluated on downstream gene sequence classification tasks. We perform extensive evaluation in both few-shot (five datasets) and full dataset settings (Genomic Understanding Evaluation benchmark consisting of 27 tasks). Our findings reveal that CM-GEMS outperforms state-of-the-art models (DNABert-2, Nucleotide transformer, DNABert) trained at 120K steps, achieving similar results in just 10K and 1K steps. We also demonstrate that Curriculum-Learned LOGO (a 2-layer DNABert-like model) can achieve nearly 90% of the state-of-the-art model performance of 120K steps. We will make the models and codes publicly available on GitHub.

This work presents a dynamic vocabulary adaptation strategy, MEDVOC, for fine-tuning pre-trained language models (PLMs) like BertSumAbs, BART, and PEGASUS for improved medical text summarization. In contrast to existing domain adaptation approaches in summarization, MEDVOC treats vocabulary as an optimizable parameter and optimizes the PLM vocabulary based on fragment score conditioned only on the downstream task’s reference summaries. Unlike previous works on vocabulary adaptation (limited only to classification tasks), optimizing vocabulary based on summarization tasks requires an additional, extremely costly intermediate fine-tuning step on large summarization datasets. To that end, our novel fragment score-based hyperparameter search very significantly reduces this fine-tuning time—from 450 days to less than 2 days on average. Furthermore, while previous works on vocabulary adaptation are often primarily tied to single PLMs, MEDVOC is designed to be deployable across multiple PLMs (with varying model vocabulary sizes, pre-training objectives, and model sizes) – bridging the limited vocabulary overlap between the biomedical literature domain and PLMs. MEDVOC outperforms baselines by 15.74% in terms of Rouge-L in zero-shot setting and shows gains of 17.28% in high Out-Of-Vocabulary (OOV) concentrations. Our human evaluation shows MEDVOC generates more faithful medical summaries (88% compared to 59% in baselines).

GPT-4 demonstrates high accuracy in medical QA tasks, leading with an accuracy of 86.70%, followed by Med-PaLM 2 at 86.50%. However, around 14% of errors remain. Additionally, current works use GPT-4 to only predict the correct option without providing any explanation and thus do not provide any insight into the thinking process and reasoning used by GPT-4 or other LLMs. Therefore, we introduce a new domain-specific error taxonomy derived from collaboration with medical students. Our GPT-4 USMLE Error (G4UE) dataset comprises 4153 GPT-4 correct responses and 919 incorrect responses to the United States Medical Licensing Examination (USMLE) respectively. These responses are quite long (258 words on average), containing detailed explanations from GPT-4 justifying the selected option. We then launch a large-scale annotation study using the Potato annotation platform and recruit 44 medical experts through Prolific, a well-known crowdsourcing platform. We annotated 300 out of these 919 incorrect data points at a granular level for different classes and created a multi-label span to identify the reasons behind the error. In our annotated dataset, a substantial portion of GPT-4’s incorrect responses is categorized as a "Reasonable response by GPT-4," by annotators. This sheds light on the challenge of discerning explanations that may lead to incorrect options, even among trained medical professionals. We also provide medical concepts and medical semantic predications extracted using the SemRep tool for every data point. We believe that it will aid in evaluating the ability of LLMs to answer complex medical questions. We make the resources available at https://github.com/roysoumya/usmle-gpt4-error-taxonomy .

Large-scale language models such as DNABert and LOGO aim to learn optimal gene representations and are trained on the entire Human Reference Genome. However, standard tokenization schemes involve a simple sliding window of tokens like k-mers that do not leverage any gene-based semantics and thus may lead to (trivial) masking of easily predictable sequences, and subsequently inefficient Masked Language Modeling (MLM) training. Therefore, we propose a novel masking algorithm, GENEMASK, for MLM training of gene sequences, where we randomly identify positions in a gene sequence as mask centers and locally select the span around the mask center with the highest Normalized Pointwise Mutual Information (NPMI) to mask. We observe that in the absence of human-understandable semantics in the genomics domain (in contrast, semantic units like words and phrases are inherently available in NLP), GENEMASK-based models substantially outperform the SOTA models (DNABert and LOGO) over four benchmark gene sequence classification datasets in five few-shot settings (10 to 1000-shot). More significantly, the GENEMASK-based DNABert model is trained for less than one-tenth of the number of epochs of the original SOTA model. We also observe a strong correlation between top-ranked PMI tokens and conserved DNA sequence motifs, which may indicate the incorporation of latent genomic information. The codes (including trained models) and datasets are made publicly available at unmapped: uri https://github.com/roysoumya/GeneMask.

Clinical trials are an essential source of information for practicing Evidence-Based Medicine because they help to determine the efficacy of newly developed treatments and drugs. However, most of the existing trial search systems focus on a specific disease (e.g., cancer) and utilize disease-specific knowledge bases that hinder the adaptation of such methods to new diseases. In this work, we overcome both limitations and propose a graph-based model that explores both clinical trials and the Pubmed databases to alleviate the shortage of relevant clinical trials for a query. We construct a large heterogeneous graph (750K nodes and 1.2 Million edges) made of clinical trials and Pubmed articles linked to clinical trials. As both the graph edges and nodes are labeled, we develop a novel metapath-based similarity search (MPSS) method to retrieve and rank clinical trials across multiple disease classes. We primarily focus on consumers and users that do not have any prior medical knowledge. As there are no multiple disease-wide trial search evaluation datasets, we contribute a high-quality, well-annotated query-relevant trial set comprising around 25 queries and, on average, approximately 95 annotated trials per query. We also perform a detailed evaluation of MPSS on the TREC Precision Medicine Benchmark Dataset, a disease-specific clinical trial search setting. We make all the codes and data publicly available at https://github.com/roysoumya/MPSS-clinical-trial-search.

Online medical forums have become a predominant platform for answering health-related information needs of consumers. However, with a significant rise in the number of queries and the limited availability of experts, it is necessary to automatically classify medical queries based on a consumer’s intention, so that these questions may be directed to the right set of medical experts. Here, we develop a novel medical knowledge-aware BERT-based model (MedBERT) that explicitly gives more weightage to medical concept-bearing words, and utilize domain-specific side information obtained from a popular medical knowledge base. We also contribute a multi-label dataset for the Medical Forum Question Classification (MFQC) task. MedBERT achieves state-of-the-art performance on two benchmark datasets and performs very well in low resource settings.