I'm a PhD researcher in the Department of Bioinformatics at the University of North Bengal, working with Dr. Chiranjib Sarkar at the Computational Systems Biology Lab. I train deep learning models on protein sequences. Much of that work has been on predicting protein-protein interactions with attention-based architectures, using transfer learning from pretrained protein language models, and more recently on the generative side: designing protein binders with autoregressive and diffusion-based models. I also build web platforms that make these trained models usable for inference and for visualizing the interaction networks they produce. My current interests include LLM fine-tuning, reinforcement learning for sequence optimization, and discrete diffusion for biological sequence generation. I am open to research collaborations in these areas, so feel free to get in touch.

Highlights

Deep-Interact Studio workflow

Deep-Interact Studio

No-code platform to design, train, compare, and interpret deep learning models for biomolecular interaction prediction.

 MoE-Bind architecture

MoE-Bind

Autoregressive, sequence-only protein binder generator combining Multi-head Latent Attention with a sparse Mixture-of-Experts.

Publications

Sarkar, D., & Sarkar, C. (2026). ARACoFusion: Uncertainty-aware calibrated deep learning for protein-protein interaction network prediction in Arabidopsis thaliana. bioRxiv, 2026.05.22.727120.

DOI: https://doi.org/10.64898/2026.05.22.727120

Sarkar, D., & Sarkar, C. (2025). AttnSeq-PPI: Enhancing protein-protein interaction network prediction using transfer learning-driven hybrid attention. Biochimica et Biophysica Acta (BBA)-Proteins and Proteomics, 141102.

DOI: https://doi.org/10.1016/j.bbapap.2025.141102

Package 'EGRNi': Gene Regulatory Network Inference

Sarkar, C.; Sarkar, D.; Parsad, R.; Mishra, D.

CRAN (R package), 2022

cran

Software

Deep-Interact Studio : Interactive web platform for building, training, and comparing custom deep learning models for protein-protein, drug-target, RNA-protein, and protein-DNA interaction prediction, with integrated interpretability.

AttnSeq-PPI : Sequence-only protein-protein interaction prediction using hybrid attention and transfer learning from pretrained protein language models.

ARACoFusion-PPI : Arabidopsis thaliana-specific PPI prediction and interaction network analysis tool.

2026

  • Aug

    Wrote a Medium walkthrough on pre-training an autoregressive Mixture-of-Experts protein language model from scratch.

  • Jul

    Deep-Interact Studio preprint is out on bioRxiv, a no-code platform for building deep learning models of biomolecular interactions.

  • Jun

    MoE-Bind preprint released: sequence-only protein binder generation with sparse Mixture-of-Experts.

  • May

    ARACoFusion preprint released: uncertainty-aware calibrated deep learning for PPI networks in Arabidopsis thaliana.

  • Apr

    Received a Google TRC award for free access to Cloud TPUs (v4/v5e/v6e).

2025

  • Nov

    AttnSeq-PPI published in Biochimica et Biophysica Acta (BBA) - Proteins and Proteomics.

  • Sep

    Awarded GPU cloud compute (NVIDIA L4) under the IndiaAI Compute Initiative.

  • Mar

    Oral presentation at Anusandhan 2025, Bose Institute, Kolkata.

Education

Ph.D. in Bioinformatics

University of North Bengal, India

Advisor: Dr. Chiranjib Sarkar · Deep learning for PPI network prediction

M.Sc. in Botany (Biochemistry)

University of North Bengal, India

First Class · 77.25%

B.Sc. (Hons.) in Botany

Ananda Chandra College, University of North Bengal

First Class · 63.50%

* expected

Experience

PhD Research Scholar

Dept. of Bioinformatics, University of North Bengal

Deep learning for sequence-based PPI prediction; attention-based hybrid models with pretrained protein language models; deployed web platforms for model inference.

Skills

Programming
Python, PyTorch, Hugging Face Transformers, R (CRAN package development), C (basic)
Deep Learning
Transformers, attention mechanisms, sequence modeling, transfer learning with pretrained language models, fine-tuning (LoRA / QLoRA), LLMs, RAG
Generative AI
Autoregressive sequence generation, diffusion-based approaches, probabilistic sequence modeling
Bioinformatics
Sequence-based protein–protein interaction prediction, protein language models, generative modeling of biological sequences
Scalable Training
GPU-based training, large-scale sequence datasets, parallel training on HPC environments
Data & Viz
NumPy, Pandas, Matplotlib, Seaborn
Deployment & Web
Docker, Flask, FastAPI, Linux server deployment, model-backed web apps, React, Vite, Tailwind CSS, Git

Awards & Qualifications

  • CSIR-NET Junior Research Fellowship (JRF) · Life Sciences · June 2021 · All India Rank 216
  • GATE Life Sciences (XL) · Qualified · 2022

Grants & Computing Resources

  • IndiaAI Compute Initiative (2025)

    Awarded GPU cloud compute (NVIDIA L4) under the IndiaAI Mission for the project Deep learning-based framework for protein-protein interaction network prediction (Project ID: P1-S2025070964, Sep 2025 - Aug 2026). Funded under CSIR-UGC SRF Fellowship.

  • Google TRC (TPU Research Cloud) Award (2025)

    Granted free access to Google Cloud TPUs (v4, v5e, v6e) for machine learning research.

Talks & Presentations

A Webtool for Transfer Learning Based Protein-Protein Interaction Network Prediction Using Hybrid Attention

Anusandhan 2025, WILEY Sponsored Oral Presentation (SARANSH)

Bose Institute, Kolkata, India · March 7, 2025 Oral

Writing

Pre-train Your First Autoregressive Mixture-of-Expert Based Protein Language Model : a hands-on walkthrough of pre-training an autoregressive, Mixture-of-Experts protein language model from scratch.