Research & publications

Advancing the science of programmable medicine.

We publish, open-source, and collaborate: peer-reviewed papers and preprints in biomedical AI, molecular design, drug-target interaction, and scientific evaluation, including work by team members before DeepBio Scientific.

Read the papers →Collaborate with us

mRNA foundation models
Drug-target interaction
Molecular design

20Selected publications

5Research areas

OpenCode and data where we can

Research areas

5 programs

01 · FOUNDATION3 papers

mRNA & molecular foundation models

Sequence- and structure-aware models that learn the language of mRNA and small molecules.

02 · BINDING5 papers

Drug–target interaction

Interpretable models that predict how molecules bind their protein targets.

03 · ARCHITECTURE3 papers

Language-model architecture

How bidirectionality, attention, and generation shape what language models learn.

04 · EFFICIENCY5 papers

Efficient & adaptable LLMs

Parameter-efficient fine-tuning and mixture-of-experts that adapt large models cheaply.

05 · APPLIED4 papers

Applied & trustworthy LLMs

LLMs applied to forecasting, user modeling, and multi-agent systems, and evaluated for fairness.

Sequence becomes medicine

Selected publications and preprints

A designed mRNA sequence encapsulated in a lipid nanoparticle, cutaway view

Formulation · lipid nanoparticlePeer-reviewed venues and preprint servers

Publications

20 papers

NeurIPS ’25Equi-mRNA: Protein-Translation Equivariant Encoding for mRNA Language ModelsFirst codon-level equivariant mRNA language model — ~10% better accuracy, ~4× more realistic constructs, ~28% better functional-property preservation.Yazdani-Jahromi M., Khodabandeh Yalabadi A., Garibay O.O.Paper →arXiv ’26Smarter Saboteurs, Better Fixers: Scaling & Security in Linear Multi-Agent WorkflowsHow scaling and adversarial pressure shape reliability in linear multi-agent LLM workflows.McAllister T., Abdidizaji S., Garibay I., Garibay O.O.Paper →arXiv ’26Monkey Jump: MoE-Style PEFT for Efficient Multi-Task LearningMixture-of-experts-style parameter-efficient fine-tuning for multi-task learning.Prottasha N.J., Kowsher M., Yu C.N., Chen C., Garibay O.Paper →arXiv ’26LiME: Lightweight Mixture of Experts for Efficient Multimodal Multi-Task LearningA lightweight mixture-of-experts for efficient multimodal, multi-task learning.Kowsher M., Mansoor H., Prottasha N.J., Garibay O., Zhu V., Ji Z., Chen C.Paper →arXiv ’26ConRetroBert: EMA-Stabilized Dual Encoders for Template-Based Single-Step RetrosynthesisDual-encoder retrosynthesis with EMA-stabilized training for template-based single-step prediction.Basher M.J.I., Khodabandeh Yalabadi A., Garibay I., Garibay O.O.Paper →HCII ’25Evaluating Fairness and Bias in Large Language Models for Tabular DataMeasuring fairness and bias when LLMs are applied to structured, tabular data.Tayebi A., Garibay O.O.Paper →arXiv ’25User Profile with Large Language Models: Construction, Updating, and BenchmarkingConstructing, updating, and benchmarking user profiles with large language models.Prottasha N.J., Kowsher M., Raman H., Anny I.J., Bhat P., Garibay I., Garibay O.Paper →TRL Workshop ’25LLM-Mixer: Multiscale Mixing in LLMs for Time-Series ForecastingMultiscale mixing that adapts LLMs to time-series forecasting.Kowsher M., Sobuj M.S.I., Prottasha N.J., Alanis E.A., Garibay O., Yousefi N.Paper →NAACL Findings ’25BnTTS: Few-Shot Speaker Adaptation in a Low-Resource SettingFew-shot speaker adaptation for text-to-speech in low-resource languages.Basher M.J.I., Kowsher M., Islam M.S., Nandi R.N., Prottasha N.J., Menon M.H.Paper →arXiv ’25PEFT A2Z: Parameter-Efficient Fine-Tuning Survey for Large Language and Vision ModelsA comprehensive survey of parameter-efficient fine-tuning across language and vision models.Prottasha N.J., Chowdhury U.R., Mohanto S., Nuzhat T., Sami A.A., Ali M.S.Paper →ACL ’25Predicting Through Generation: Why Generation Is Better for PredictionEvidence that generative objectives beat discriminative ones for prediction tasks.Kowsher M., Prottasha N.J., Bhat P., Yu C.N., Soltanalian M., Garibay I., Garibay O.Paper →NAACL ’25Does Self-Attention Need Separate Weights in Transformers?A shared-weight self-attention that trims parameters without sacrificing accuracy.Kowsher M., Prottasha N.J., Yu C.N., Garibay O., Yousefi N.Paper →arXiv ’25How Bidirectionality Helps Language Models Learn Better via Dynamic Bottleneck EstimationDynamic bottleneck estimation quantifying the advantage bidirectionality gives representation learning.Kowsher M., Prottasha N.J., Xu S., Mohanto S., Chen C., Garibay O.Paper →Bioinf. Adv. ’25BoKDiff: Best-of-K Diffusion Alignment for Target-Specific 3D Molecule GenerationBest-of-K diffusion alignment for target-specific 3D molecule generation.Khodabandeh Yalabadi A., Yazdani-Jahromi M., Garibay O.O.Paper →Sci. Reports ’24Parameter-Efficient Fine-Tuning of LLMs Using Semantic Knowledge TuningFine-tuning LLMs with meaningful semantic prompts instead of random tokens.Prottasha N.J., Mahmud A., Sobuj M.S.I., Bhat P., Kowsher M., Yousefi N.Paper →J. Biomol. Struct. Dyn. ’24Examining Sialic-Acid Derivatives as Potential Inhibitors of the SARS-CoV-2 Spike RBDComputational screening of sialic-acid derivatives against the SARS-CoV-2 spike receptor-binding domain.Banerjee T., Gosai A., Yousefi N., Garibay O.O., Seal S., Balasubramanian G.Paper →RECOMB ’24FragXsiteDTI: Revealing Responsible Segments in Drug–Target Interaction with Transformer-Driven InterpretationInterpretable transformer that surfaces the fragments and sites responsible for a drug–target interaction.Khodabandeh Yalabadi A., Yazdani-Jahromi M., Yousefi N., Tayebi A., Abdidizaji S.Paper →Briefings in Bioinf. ’23BindingSite-AugmentedDTA: A Next-Generation Pipeline for Interpretable Prediction in Drug RepurposingBinding-site-aware augmentation that makes drug-repurposing predictions interpretable.Yousefi N., Yazdani-Jahromi M., Tayebi A., Kolanthai E., Neal C.J.Paper →Molecules ’22UnbiasedDTI: Mitigating Real-World Bias of Drug–Target Interaction PredictionDeep ensemble-balanced learning that reduces real-world bias in DTI prediction.Tayebi A., Yousefi N., Yazdani-Jahromi M., Kolanthai E., Neal C.J., Seal S.Paper →Briefings in Bioinf. ’22AttentionSiteDTI: An Interpretable Graph-Based Model for Drug–Target Interaction PredictionGraph-based DTI model that borrows NLP sentence-level relation classification for interpretability.Yazdani-Jahromi M., Yousefi N., Tayebi A., Kolanthai E., Neal C.J., Seal S.Paper →

Open source

We release models, benchmarks, and tooling to move the whole field of mRNA design forward.

Browse the repositories →

Collaborate with us

We partner with academic labs and clinical groups on foundational problems in mRNA science.

Start a collaboration →