Theodor Spiro
ORCID 0009-0004-5382-9346 · github.com/mool32 · mool32.github.io · linkedin.com/in/theodorspiro
Profile
Independent computational researcher with a biophysics background (Lomonosov Moscow State University). I work on substrate-independent organizing principles across cells, aging tissues, and neural networks — applying one statistical apparatus across biological and engineered systems. Current center of the work is perceptome, a framework and Python toolkit treating cellular signaling pathways as a perceptual repertoire. Alongside: cross-population aging biomarkers (ECG, EEG, transcriptome) and the comparative biology of neural networks (distribution of fitness effects, epistasis). Methodology inherited from experimental physics: preregistration, bitwise-reproducible pipelines, bootstrap uncertainty quantification, and honest negative results.
Software & tools
perceptome (2026) — Python toolkit for cellular perception analysis. 44 signaling modules; 9-PC eigenspace built from 154 Human Protein Atlas cell types; capacity-floor predictor; validity scorecard with three null controls; 8-cell cancer-attractor reference (11 cancers from 11 organ systems converge toward it during transformation, independently replicated on an external HCC cohort). Zenodo 10.5281/zenodo.20113468 · 73/73 tests passing.
Publications & preprints
Preprints (arXiv / bioRxiv / Zenodo):
- Spectral exponents of the twelve-lead ECG reveal the anatomy of cardiac conduction disorders and a bifurcation between aging and disease. bioRxiv (submitted), 2026. 412,730 recordings across three continents; CLBBB vs CRBBB AUC ≈ 0.98 cross-population. Zenodo 10.5281/zenodo.19945065.
- Transcriptomic noise accumulates within tissue identity across human aging. bioRxiv (post-review v4), 2026. GTEx v8 + Tabula Muris Senis + Calico rat + macaque atlas; aging as systemic noise, not selective accumulation. Zenodo 10.5281/zenodo.19944444.
- Universal statistical signatures of evolution in artificial intelligence architectures. arXiv:2604.10571, 2026. 935 ablation experiments across 161 publications; substrate-independent heavy-tailed DFE.
Under review:
- Waveform asymmetry as a biomarker of neural aging. Frontiers in Aging Neuroscience, 2026. LEMON (N=215) + Dortmund Vital Study (N=608) + 208-subject 5-year longitudinal. Zenodo 10.5281/zenodo.19912202.
- Functional differentiation generates universal fitness-effect distributions in neural networks. Under review, NeurIPS 2026 / ICML Mechanistic Interpretability Workshop 2026. Pythia 410M, 1,584 controlled ablations across 8 checkpoints.
Manuscripts in preparation:
- Clonal crystallization as a shared signature of bone-marrow aging and neural-network training. Cross-substrate (Gini, eff_N) framework; caloric restriction rescues 64% of the aging drift in rat bone marrow.
- Temporal architecture of signaling oscillations predicts cancer gene function across pathways. Rise/recovery temporal classification; OR = 27.5, p = 3.6 × 10⁻⁹ across 14 pathways.
- Negative feedback loop architecture as a modular predictor of cancer vulnerability. 128 NFLs from KEGG; 59-fold CGC enrichment (p = 9 × 10⁻⁴⁴); Irreversible Authority metric (ρ = 0.83).
- The Oracle’s Fingerprint: correlated AI forecasting errors and the limits of bias transmission. GPT-4o / Claude / Gemini error correlation r = 0.78 on 568 Metaculus forecasts.
- 21,000 attempts to think differently: a Russian adaptation of the Divergent Association Task. N = 21,159; Cronbach’s α = 0.899; live instrument deployed.
- Self-specific representations localize in emergent attention heads. Pilot: 29 active meta-heads, all Δ_self > Δ_cross (binomial p < 10⁻⁸).
Technical reports:
- EEG-based comprehension detection: a multi-dataset non-replication. Honest negative result; 18 metrics, 5 datasets, 126 subjects.
Research experience
Independent Researcher — affiliated with Vaika Inc. · 2020 – present Independent computational research program across cellular perception, aging biomarkers, comparative biology of neural networks, and AI-collaborative research methodology. Recent concentration on the perceptome framework and on single-cell / transcriptomic analysis; earlier work in computational neuroscience (EEG) and systems biology. All projects preregistered where inferential; full code and data released on GitHub and Zenodo.
B.Sc. / M.Sc. research — Moscow State University, Faculty of Physics · 2018 – 2022 Supervisor: Prof. L. Yakovenko. Agent-based / cellular-automata modeling of the cellular response to pro-inflammatory stimuli (TLR4/TLR6 → NF-κB → TNF → apoptosis), identifying key parameters of the innate-immunity signaling pathway. Presented at the Lomonosov-2021 Scientific Conference.
Research methodology
Developer of an AI-collaborative research methodology framework (mool32.github.io/methodology) — preregistration discipline, sign-convention locks, locked-vs-working artifacts, reproducibility standards, and AI-friendly publishing — applied as worked examples across the portfolio.
Education
M.Sc. Biophysics (coursework completed) — Lomonosov Moscow State University, Faculty of Physics · 2021 – 2022 Computer simulation in biology; physics of biopolymers; physicochemical kinetics; magnetic radio-spectroscopy in biology and medicine. Thesis defense not completed due to relocation from Russia in 2022.
B.Sc. Biophysics — Lomonosov Moscow State University, Faculty of Physics · 2014 – 2021
Lyceum 1525, Physics & Mathematics — Moscow · 2010 – 2014
Teaching
Eleven years teaching mathematics and physics (ages 11–25), including gifted and olympiad-oriented students, across selective international programs, online schools, and private practice.
- Le Sallay Academy — STEM Teacher & Curriculum Designer · 2025 – 2026. Selective international program for gifted students (France residential + online); designed and delivered the full mathematics, physics, and earth-science curriculum.
- BrainyBara — Co-Founder & Lead Teacher · 2022 – present. Mathematics, physics, and critical-thinking programs (Israel); Bagrut and Edexcel A-Level prep; built AI-powered teaching tools.
- Foxford Online School — Mathematics & Physics Teacher · 2015 – 2018 (Moscow). 12+ courses developed.
- Moscow State University — Instructor, Advanced Mathematics · 2018 – 2020.
- Private tutoring & methodology · 2014 – 2022.
Professional development
- Evidence-Based Teaching Practices — Harvard University BOK Center · 2022 – 2024
- Psychology of Development and Learning — MSUPE, Moscow · 2020
- Learning to Teach Online — University of New South Wales · 2020
Grants & honors
- Emergent Ventures grant — 2026
Technical skills
- Programming: Python (NumPy, Pandas, SciPy, scikit-learn, matplotlib, PyTorch), Git, Bash, LaTeX, SQL basics
- Methods: preregistered experimental design, bootstrap uncertainty quantification, distribution fitting with AIC comparison, dimensionality reduction, time-series and signal analysis, agent-based modeling
- Data domains: single-cell & bulk RNA-seq, EEG / ECG signals, neural-network interpretability, large-scale forecasting data
- AI-assisted research: LLM-integrated analysis pipelines under an explicit preregistration / reproducibility framework
Languages
Russian (native) · English (C1; all research published in English) · Hebrew (B1)