A = U Σ Vᵀ
initializing membrane potential ...
Computational & Systems Biology

Bhuwan Sharma

Spectral Methods · Network Computation · Systems Biology

I work on grounded approaches for modeling biological systems - extracting structure from high-dimensional data through decomposition, graph theory, and principled computation.

Gene regulatory network · Live eigendecomposition
02 -

Background

Biological data is high-dimensional, structured, and noisy. My work centers on extracting meaningful structure from it using statistical reasoning and graph-based computation.

Core Interests

  • Spectral analysis of biological networks - Laplacian eigenmaps, Fiedler vectors, community detection
  • Dimensionality reduction with algebraic structure - SVD, principal component networks, manifold geometry
  • Ontology-driven feature construction - GO annotations → learnable numerical representations
  • Efficient computation for high-dimensional systems - GPU acceleration, sparse matrix methods, scalable pipelines

Rather than stacking models, I focus on understanding the geometry and algebra behind them.

Gene expression signaling · simulated

Education

2023–2025
M.Sc. Bioinformatics
University of North Bengal
Computational systems biology, ML-driven analysis of high-dimensional data
2020–2023
B.Sc. Zoology (Honours)
Kalimpong College - University of North Bengal
CGPA: 7.64 · Molecular biology, biochemistry and genetics foundation
03

Research Directions

Current Focus

My research is oriented toward principled and interpretable computational biology:

  • λ Spectral approaches for biological network inference
  • Efficient computation in high-dimensional systems
  • Structured integration of biological knowledge into learnable representations

The long-term direction: computation that respects the algebraic structure of biology while remaining scalable and interpretable.

04 -

Projects

Statistical Foundations

Stats

Foundational statistical analysis and visualization - understanding data behavior before applying complex models.

Distribution analysis Regression Exploratory interpretation Reproducible workflows

GPU-Accelerated Principal Component Network

PCnet_GPUacc

Accelerating principal component–based network computations using GPU resources. Mathematical structure meets computational scalability.

Eigen-decomposition Matrix factorization Network inference CUDA optimization

Ontology-Based Feature Structuring

Integrating Gene Ontology annotations into structured computational formats - bridging biological semantics with numerical modeling.

Gene–ontology mappings Hierarchical features High-dim annotation
05 -

Technical Expertise

Programming Languages

Python R Bash C Java

Bioinformatics

limma DESeq2 clusterProfiler biomaRt

Machine Learning

PyTorch scikit-learn Transformers PyTorch Geometric

Systems & Tools

Linux Git Conda CUDA Jupyter
06 -

Approach

01

Understand the root structure

Before reaching for algorithms - ask what is the underlying form??
The answer shapes everything that follows.

02

Reduce to what matters

High-dimensional biological data carries noise alongside signal. Spectral methods, dimension reduction, and structured feature construction isolate what is biologically meaningful.

03

Preserve biological meaning

Mathematical abstraction should serve biology, not replace it. Every transformation should be interpretable - a principal component should map back to a gene set, a cluster to a phenotype.

04

Build toward impact

The end goal is not a model - it is insight. Biomarker discovery, disease understanding, drug target identification. Computation should ultimately serve human health.

07

Spectral Embedding

"A graph Laplacian's eigenvectors encode geometry.

Communities emerge from spectrum, not from labels."

The interactive below demonstrates how spectral methods extract geometric structure from connectivity. The network's Laplacian matrix is decomposed - its eigenvectors become coordinates in embedding space, revealing clusters that topology alone implies.

3D Spectral Embedding drag to rotate
Network Topology hover nodes
Laplacian Spectrum
Community A Community B Community C Bridge nodes
L = D − A  →  Lv = λv  →  embedding: [v₂, v₃, v₄]

Biology is not a benchmark dataset - it is a system where every gene, every interaction, every annotation carries meaning shaped by evolution. The most impactful computational biology comes not from chasing the newest architecture, but from deeply understanding the problem's structure.

A well-chosen decomposition reveals more than a thousand gradient steps. I prioritize clarity in formulation, disciplined computation, and structured reasoning - because the goal is not just to predict, but to understand.

08 -

Contact

Interested in collaborative research or have a project in mind?

Open for collaboration in:

  • Spectral methods in systems biology
  • Network-based inference and prediction
  • Multi-omics data integration
  • Machine and Deep Learning for biology