Tuba Aksoy, PhD

Computational Biology · Neuroscience · Statistical Genetics · Machine Learning

Computational biologist with a PhD in quantitative systems biology and neuroscience and an undergraduate background in physics. I develop statistical and mechanistic models of biological systems: single-neuron dynamics, spiking networks, and statistical genetics, with a focus on methods that transfer across domains.

GitHub LinkedIn Email

Research Interests

Experience

Bioinformatics Scientist - Statistical Genetics & Clinical Cohorts

  • Led epigenome-wide association analysis over ~4.7 billion methylation measurements (860K EPIC probes × 5K samples; ~2,700 participants, two visits), designing a two-stage mixed-effects framework to handle repeated measures and batch structure that standard EWAS tooling couldn't express.
  • Built a 15-model sensitivity ladder with per-model genomic-inflation (λ) monitoring to separate robust associations from confounder-driven ones; restructured ~800K per-probe fits into chunked, checkpointed concurrent batch jobs, restartable rather than all-or-nothing.
  • Modeled cerebrovascular MRI phenotypes with GENESIS variance-component models (kinship, household, census-block random effects); implemented methylation risk scores for lead exposure from published CpG panels.

Graduate Research Assistant - Bioinformatics & Multi-Omics

  • Led an end-to-end multi-omic study of radiation neurotoxicity in mouse hippocampus and PFC across snRNA-seq and WGBS, from treatment and library prep to full computational analysis.
  • Identified TTR as the most dysregulated gene across all hippocampal cell types (snRNA-seq), with convergent WGBS evidence placing it beside the most differentially methylated tile, pointing to epigenetic regulation upstream of transcription.
  • Designed a custom framework linking 1000-bp methylation tiles to gene expression by genomic distance, revealing a distance-dependent methylation-to-expression relationship consistent across cell types.

Machine Learning - Connectomics & Neural Systems

  • Built an ML pipeline predicting synapse formation from the MICrONS connectome (186K axo-dendritic instances), reaching 0.789 balanced accuracy on a severely imbalanced dataset.
  • Engineered 50+ multimodal features (functional-embedding cosine similarity, morphological embeddings, spatial geometry), lifting balanced accuracy from 0.74 to 0.79; found postsynaptic features the strongest predictors, a biologically interpretable result.

Guest Lecturer & Curriculum Developer - Theoretical Neuroscience

  • Delivered three graduate lectures on neuronal membrane dynamics (Hodgkin-Huxley, Connor-Stevens, cable theory) to a mixed CS/neuroscience audience; developed and graded assessments.

Graduate Research Assistant - Mechanistic Modeling & Dynamical Systems

  • Built ODE-based recurrent spiking networks adding calcium / Ca-activated cation channels to leaky integrate-and-fire neurons, resolving a bistability limit and extending trainable timescales from 900 ms to 16,000 ms (18×). First-author, J. Comput. Neurosci. 2022.
  • Developed a custom grid-search framework to explore 10+ conductance and synaptic parameters, mapping the relationship between synaptic strength and decay time that governs long-timescale stability.
  • Designed and deployed NeuroTune, an interactive neuron-model simulator and teaching platform.

Projects & Software

AIF Spiking Neural Network

Publication · J Comput Neurosci 2022

Recurrent spiking network of active integrate-and-fire neurons carrying calcium and calcium-activated cation conductances. The added conductances sustain long-lasting activity at realistic firing rates without slow NMDA synapses, and let the network learn intervals an order of magnitude longer than leaky integrate-and-fire networks.

Paper →  ·  ModelDB →  ·  Code →

NeuroTune

Interactive simulator · teaching site

An interactive single-neuron simulator and teaching resource covering the Leaky Integrate-and-Fire, Hodgkin-Huxley, and Connor-Stevens models. Built in MATLAB (App Designer) and ported to a Python/Streamlit web app.

Website →  ·  Launch simulator →

MRI Brain-Age Prediction

Machine learning · neuroimaging

Predicting chronological age from structural MRI (OASIS-1, Nilearn) with ridge regression and age-bias correction, then testing whether the Brain-Age Gap tracks cognitive scores. MAE 9.2 years, R² 0.77.

GitHub →

ML Foundations

Applications & theory · R, Python

Machine learning from first principles: KNN, regularization and feature selection, cross-validation, clustering/PCA and graphical models, neural nets and tree ensembles. Each topic pairs implementations with hand-written mathematical derivations.

GitHub →

Publications

  1. Aksoy, T. & Shouval, H. Z. (2022). Active intrinsic conductances in recurrent networks allow for long-lasting transients and sustained activity with realistic firing rates as well as robust plasticity. Journal of Computational Neuroscience, 50(1), 121-132.
    DOI · Free full text (PMC) · Model (ModelDB) · Code
  2. Aksoy, T. (2024). Early onset Alzheimer's disease markers in mouse hippocampus unveiled by single-cell transcriptomic analysis following cranial radiotherapy. PhD dissertation, UTHealth & MD Anderson GSBS.
    DigitalCommons@TMC
  3. Aksoy, T. et al. Fractionated cranial radiation induces cell-type-specific transcriptional dysregulation with associated DNA methylation changes in the mouse brain. Clinical and Translational Radiation Oncology. Manuscript under review.

Background

Education

  • PhD, Quantitative Systems Biology & Neuroscience (dual degree)
    UT MD Anderson Cancer Center & UTHealth Graduate School of Biomedical Sciences, Houston, TX · GPA 4.0
  • Pre-doctoral graduate studies
    UT MD Anderson Cancer Center & UTHealth Graduate School of Biomedical Sciences, Houston, TX
  • BS, Physics (minor in Astrophysics)
    Boğaziçi University, Istanbul, Turkey

Skills

  • Programming: Python, R, MATLAB, SQL, Bash; HPC, Git
  • Statistical genetics: EWAS, mixed-effects models, PRS, epigenetic age, PCA/population structure, FDR
  • Genomics: snRNA-seq, WGBS, multi-omic integration, differential expression, cell-type deconvolution
  • Machine learning: supervised & unsupervised, feature engineering, XGBoost, elastic net/lasso, SVM, deep nets (PyTorch, Keras)
  • Modeling: ODE-based systems, mechanistic & dynamical systems, parameter estimation, sensitivity analysis