Vijay Narayan
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Research Interests
Vijay Narayan has a total of 12 co-authors. Here are the top co-authors by number of shared publications:
- Daniel S. Berman
Johns Hopkins Applied Physics Laboratory, 11100 Johns Hopkins Rd., Laurel, MD 20723, USA
2 shared publications - Amanda Ernlund
Johns Hopkins University Applied Physics Laboratory
1 shared publication - Brian Merritt
Johns Hopkins Applied Physics Laboratory, 11100 Johns Hopkins Rd., Laurel, MD 20723, USA
1 shared publication - Brian Merritt
Johns Hopkins University Applied Physics Laboratory
1 shared publication - Dana Udwin
Johns Hopkins Applied Physics Laboratory, 11100 Johns Hopkins Rd., Laurel, MD 20723, USA
1 shared publication - Daniel Jimenez
Johns Hopkins University Applied Physics Laboratory
1 shared publication
Latest Publications
- Open Access
MicrobeRT: A Framework and Suite of Fine-tuned Language Models for Genomic Sequences
Accurate and reproducible metagenomic analysis is essential for effective infectious disease surveillance. Targeted diagnostics such as culturing, antigen assays, and multiplex PCR panels identify only predefined organisms and can fail to identify atypical or novel infections. Shotgun metagenomics offers hypothesis-free detection, but adoption is limited by workflow complexity, dependence on mutable reference databases, and high computational cost. To combat these challenges, we introduce MicrobeRT, an integrated suite of genomic language models fine-tuned to classify metagenomic sequencing reads across three tasks: (i) species level taxonomic identification, (ii) antibiotic resistance gene detection, and (iii) pathogenicity assessment. The framework builds on three pre-trained genomic language models - Nucleotide Transformer (Dalla-Torre et al 2025), DNABERT (Ji et al 2021, Zhou et al 2024), and HyenaDNA (Nguyen et al 2024) - leveraging their large-scale genomic pre-training, adaptability to downstream tasks, and, in the case of HyenaDNA, ability to capture long-range sequence dependencies. The models are fine-tuned on large, curated datasets spanning microbial, viral, fungal, and human genomes, including more than 500,000 genomes and 250,000 antibiotic resistance gene sequences. To evaluate the performance of our fine-tuned genomic language models across classification tasks, we developed a benchmarking framework that compares MicrobeRT against state-of-the-art (SOTA) bioinformatics tools. Across both simulated and empirical datasets, the models achieve strong performance while reducing computational overhead. MicrobeRT is released as open-source software, enabling further fine-tuning and reproducible integration of alternative models and classifiers. We also demonstrate seamless, drop-in integration by incorporating our fine-tuned taxonomic classification models as an optional add-on within the open-source TaxTriage pipeline (Merritt et al 2026). MicrobeRT demonstrates the feasibility of three clinically relevant multi-label analyses, offering a practical path to scalable, reproducible metagenomic diagnostics and enabling broader deployment of AI/ML-enhanced infectious disease testing.
Open AccessMicrobeRT: A Framework and Suite of Fine-tuned Language Models for Genomic Sequences
Accurate and reproducible metagenomic analysis is essential for effective infectious disease surveillance. Targeted diagnostics such as culturing, antigen assays, and multiplex PCR panels identify only predefined organisms and can fail to identify atypical or novel infections. Shotgun metagenomics offers hypothesis-free detection, but adoption is limited by workflow complexity, dependence on mutable reference databases, and high computational cost. To combat these challenges, we introduce MicrobeRT, an integrated suite of genomic language models fine-tuned to classify metagenomic sequencing reads across three tasks: (i) species level taxonomic identification, (ii) antibiotic resistance gene detection, and (iii) pathogenicity assessment. The framework builds on three pre-trained genomic language models - Nucleotide Transformer (Dalla-Torre et al 2025), DNABERT (Ji et al 2021, Zhou et al 2024), and HyenaDNA (Nguyen et al 2024) - leveraging their large-scale genomic pre-training, adaptability to downstream tasks, and, in the case of HyenaDNA, ability to capture long-range sequence dependencies. The models are fine-tuned on large, curated datasets spanning microbial, viral, fungal, and human genomes, including more than 500,000 genomes and 250,000 antibiotic resistance gene sequences. To evaluate the performance of our fine-tuned genomic language models across classification tasks, we developed a benchmarking framework that compares MicrobeRT against state-of-the-art (SOTA) bioinformatics tools. Across both simulated and empirical datasets, the models achieve strong performance while reducing computational overhead. MicrobeRT is released as open-source software, enabling further fine-tuning and reproducible integration of alternative models and classifiers. We also demonstrate seamless, drop-in integration by incorporating our fine-tuned taxonomic classification models as an optional add-on within the open-source TaxTriage pipeline (Merritt et al 2026). MicrobeRT demonstrates the feasibility of three clinically relevant multi-label analyses, offering a practical path to scalable, reproducible metagenomic diagnostics and enabling broader deployment of AI/ML-enhanced infectious disease testing.
- Open Access
What You Read Is What You Classify: Highlighting Attributions to Text and Text-like Inputs
At present, there are no easily understood explainable artificial intelligence (AI) methods for discrete token inputs, like text. Most explainable AI techniques do not extend well to token sequences, where both local and global features matter, because state-of-the-art models, like transformers, tend to focus on global connections. Therefore, existing explainable AI algorithms fail by (i) identifying disparate tokens of importance, or (ii) assigning a large number of tokens a low value of importance. This method for explainable AI for tokens-based classifiers generalizes a mask-based explainable AI algorithm designed originally for images. It starts with an Explainer neural network that is trained to create masks to hide information not relevant for classification. Then, the Hadamard product of the mask and the continuous values of the classifier’s embedding layer is taken and passed through the classifier, changing the magnitude of the embedding vector but keeping the orientation unchanged. The Explainer is trained for a taxonomic classifier for nucleotide sequences and it is shown that the masked segments are less relevant to classification than the unmasked ones. This method focused on the importance the token as a whole (i.e., a segment of the input sequence), producing a human-readable explanation.
Open AccessWhat You Read Is What You Classify: Highlighting Attributions to Text and Text-like Inputs
At present, there are no easily understood explainable artificial intelligence (AI) methods for discrete token inputs, like text. Most explainable AI techniques do not extend well to token sequences, where both local and global features matter, because state-of-the-art models, like transformers, tend to focus on global connections. Therefore, existing explainable AI algorithms fail by (i) identifying disparate tokens of importance, or (ii) assigning a large number of tokens a low value of importance. This method for explainable AI for tokens-based classifiers generalizes a mask-based explainable AI algorithm designed originally for images. It starts with an Explainer neural network that is trained to create masks to hide information not relevant for classification. Then, the Hadamard product of the mask and the continuous values of the classifier’s embedding layer is taken and passed through the classifier, changing the magnitude of the embedding vector but keeping the orientation unchanged. The Explainer is trained for a taxonomic classifier for nucleotide sequences and it is shown that the masked segments are less relevant to classification than the unmasked ones. This method focused on the importance the token as a whole (i.e., a segment of the input sequence), producing a human-readable explanation.