Noelito Parina

PhD Candidate | Computational Geneticist | University of Warwick

Welcome!

I'm a computational biologist at the University of Warwick working on regulatory networks in various biological systems and how they unfold over time. My research develops transparent, parameter-aware analytical frameworks for multi-omics time-series data, particularly RNA-Seq and ATAC-Seq.

I have created SPECTRA (Systematic Parameter Exploration with Caching for Temporal RNA and ATAC), an optimisation engine that explores hundreds of thousands of analytical configurations to reveal how filtering, normalisation, clustering and enrichment decisions shape biological conclusions. SPECTRA is the first incarnation of what I call a combinatorial pipeline, a deliberate alternative to the ad-hoc, single-pipeline culture that dominates bioinformatics today.

I apply this work to decidualisation, the process by which the endometrium in the uterus prepares for embryo implantation. Reliable, reproducible analyses here matter directly for IVF outcomes and for addressing the UK's IVF "postcode lottery."

RESEARCH THEMES

Temporal Multi-omics for Diagnostics

Gene Regulatory Networks

Algorithms and Optimisation

Machine Learning and AI in ethics

CURRENT PROJECT OBJECTIVES

Temporal multi-omics for diagnostics

Understanding how gene expression and chromatin accessibility change over time is essential for decoding the regulatory dynamics that govern biological transitions. My research uses time-series RNA-seq and ATAC-seq to reconstruct the sequence of molecular events underlying processes such as decidualisation and inflammation. By capturing transient activation windows, like for example, when enhancers open before a gene is expressed, I aim to map precise regulatory trajectories that cannot be seen in static datasets.

This work uses primary samples from patients at University Hospitals Coventry and Warwickshire, supported by the Biomedical Research Unit in Reproductive Health and the Warwick Multi-omics Facility. My goal is to characterise what constitutes a receptive uterine environment for implantation.

Algorithms and optimisation

Multi-omics analysis involves dozens of analytical decisions, from filtering thresholds to clustering algorithms. Each decision shapes the biological interpretation, and in highly dynamic data even one different choice can change the story.

I have conceptualised and built combinatorial optimisation approaches that systematically explore these choices rather than relying on a single pipeline. SPECTRA evaluates over 170,000 configurations and identifies robust, reproducible analytical paths using multi-objective scoring that combines internal clustering metrics (silhouette, Calinski–Harabasz, Dunn) with biological signals (enrichment specificity, motif coherence, RNA–ATAC concordance).

Gene Regulatory Networks

RNA-seq tells us what is being expressed; ATAC-seq tells us where the regulatory machinery is open. Integrating the two gives a richer view of gene regulatory networks (GRNs), but only when both modalities are processed and aligned in ways that respect their underlying noise structure.

My work focuses on temporal GRN reconstruction by identifying which transcription factors drive which expression programmes, and when.

Machine Learning and AI ethics

Machine learning has enormous potential in multi-omics, but only if upstream pipelines are robust and interpretable. The "black-box" problem in clinical AI begins long before the model, it begins with hidden assumptions in pipeline parameters.

A long-term goal of my work is to build ML-assisted parameter selection systems: models that learn from dataset characteristics to recommend optimal configurations without exhaustive computation. This bridges methodological innovation with practical application, ensuring future AI-driven genomics is grounded in trustworthy, well-defined analysis

Why does this matter?

Bioinformatics is dominated by linear, ad-hoc pipelines built on convention rather than optimisation. Analysts default to the first reasonable settings because running alternatives is computationally expensive and operationally tedious. The result: analytical objectives stay implicit, parameter sensitivity is invisible, and biological conclusions depend on choices that are rarely justified.

This is the bioinformatics version of the "cherry-picking" problem in data science and the "black-box" challenge in machine learning. SPECTRA exposes the decision space rather than hiding it, giving each parameter choice a quantitative rationale.

What's next?

These themes come together in SPECTRA, where I investigate how analytical decisions shape biological conclusions in temporal RNA-seq and ATAC-seq. I break down the theory behind the algorithms, Mfuzz, spectral clustering, hypergeometric integration, and explain how parameter choices influence their behaviour.

-> Read about SPECTRA

Page updated

Google Sites

Report abuse