Computational Biology · Machine Learning

Satvik
Saluja

B.Tech Biotechnology + Minor in AI/ML — working at the intersection of systems biology, dynamical systems, and machine learning. Building models that think in differential equations.

Scroll

I'm a biotechnology undergraduate with a minor in AI/ML, working at the boundary where computational methods meet biological systems. My work centers on building predictive and mechanistic models of complex biological phenomena — not through curve-fitting, but through formulating the underlying dynamical structure.

My primary technical interest is in Neural Ordinary Differential Equations and state-space modeling — frameworks that treat biological processes as continuous-time dynamical systems rather than static input-output mappings. This allows models to capture memory, feedback, and temporal evolution in a physically interpretable way.

I work on systems-level biological simulation: metabolic networks, cognitive state dynamics, and pathway-level modeling. The goal is always the same — reduce complex multi-component biology to a set of coupled equations whose behavior can be analyzed, perturbed, and predicted.

On the ML side, I focus on architectures and methods where the inductive bias is grounded in domain knowledge: physics-informed networks, graph models of biological interaction networks, and hybrid mechanistic-statistical approaches.

4
Systems simulated
Neural
ODEs
Primary framework
GSoC
Targeting 2026

Selected work

001
NeuroODE
Cognitive State Simulator

Built a Neural ODE–based system to model the continuous evolution of cognitive states (attention, fatigue, stress) from EEG signals. Uses DEAP dataset features to learn dynamical behavior over time.

Approach: Learns continuous-time dynamics using dS/dt = f(S, U, θ) with torchdiffeq (dopri5 solver) instead of discrete prediction.

002
HNN Validation &
Intregrity Framework

Built an interpretable dynamical systems model using coupled differential equations (RK4) to simulate interacting cognitive states over continuous time.

Approach: Uses differential equation-based modeling with RK4 integration — learns rate of change instead of direct outputs.

003
TCA + PPP + Glycolysis
Simulation

Built a systems-level biochemical simulation integrating glycolysis, the pentose phosphate pathway, and the TCA cycle into a unified dynamical model. Tracks metabolite flux, energy balance, and cofactor dynamics using coupled differential equations.

Approach: Coupled ODE-based system with conserved ATP/NAD pools and validated through pytest-based stoichiometric testing.

004
AirAware
AQI Prediction Platform

Developed a system to estimate AQI at arbitrary geographic locations using spatial interpolation across sparse sensor networks with real-time data handling.

Approach: Uses inverse-distance weighting and spatial interpolation to predict AQI beyond available sensor data.

Technical stack

Languages
  • Python
  • C++
  • C
  • JavaScript
ML / Data
  • PyTorch
  • NumPy
  • Pandas
  • SciPy
  • torchdiffeq
Concepts
  • Neural ODEs
  • Dynamical Systems
  • Systems Modeling
  • Dynamic Programming
  • Domain Adaptation

Get in touch

Open to research collaborations, GSoC discussions, and roles at the intersection of computational biology and machine learning.