Implementation of Linear Regression and Linear Interpolation using Reaction Networks
Abstract
Statistical inference is a fundamental component of data science. In this work, we focus on two classical inference techniques: regression and interpolation. We propose a reaction-network-based framework for implementing linear regression, including both univariate and multivariate settings, as well as linear interpolation. Our approach encodes the outputs of these inference techniques in the steady-state concentrations of species within the reaction network. A key ingredient of the construction is a novel generalized division module capable of handling division involving negative numbers. We validate the proposed framework through in silico implementations on standard synthetic datasets and obtain the expected regression and interpolation outputs.
Related
- Persistence of bounded trajectories in positively transverse reaction networks
- Identifiability of a linear dynamical reaction-diffusion system with directed interactions from the second-order statistics of a single snapshot
- Layered mixed matrices and reaction networks
- Simulation-free approximate Bayesian computation for stochastic reaction networks
- Simulation-free approximate Bayesian computation for stochastic reaction networks
BibTeX
@misc{kumar2026implementation,
title = {{Implementation of Linear Regression and Linear Interpolation using Reaction Networks}},
author = {Aryan Kumar and Amey Choudhary and Jiaxin Jin and Chittaranjan Hens and Abhishek Deshpande},
howpublished = {arXiv},
year = {2026},
month = sep,
doi = {10.48550/arxiv.2606.12573},
eprint = {2606.12573},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2606.12573},
}
RIS
TY - UNPB
TI - Implementation of Linear Regression and Linear Interpolation using Reaction Networks
AU - Aryan Kumar
AU - Amey Choudhary
AU - Jiaxin Jin
AU - Chittaranjan Hens
AU - Abhishek Deshpande
PY - 2026
DA - 2026/09/19
PB - arXiv
DO - 10.48550/arxiv.2606.12573
UR - https://arxiv.org/abs/2606.12573
AB - Statistical inference is a fundamental component of data science. In this work, we focus on two classical inference techniques: regression and interpolation. We propose a reaction-network-based framework for implementing linear regression, including both univariate and multivariate settings, as well as linear interpolation. Our approach encodes the outputs of these inference techniques in the steady-state concentrations of species within the reaction network. A key ingredient of the construction is a novel generalized division module capable of handling division involving negative numbers. We validate the proposed framework through in silico implementations on standard synthetic datasets and obtain the expected regression and interpolation outputs.
ER -