Nyx Dynamics — Research Portfolio
Peer-reviewed work spanning HIV prevention implementation science, algorithmic equity, mathematical modeling of infection dynamics, and neurometabolic signal processing. All code open-source. All data archived.
Active Research
Each project includes a dedicated page with slide deck, mind map, and infographic.
Long-acting injectable PrEP demonstrates >96% HIV prevention efficacy, yet 47.1% of prescribed patients never receive their first injection. This project defines the 2–8 week “bridge period” as the critical implementation bottleneck, proposes a reconceptualized PrEP cascade, synthesizes a 21-intervention evidence library across 7 population categories, and delivers a computationally validated clinical decision support tool — validated across 21.2 million synthetic patients at UNAIDS global PrEP target scale with ±0.018 pp precision.
A mathematical and computational framework for identifying and quantifying synergistic algorithmic discrimination — where individual algorithmic features are equitable in isolation but combine multiplicatively to produce discriminatory outcomes. Proposes a recourse framework for correcting compounding bias in clinical decision support systems, with direct application to PrEP eligibility, risk stratification, and treatment recommendation algorithms.
A signal processing framework demonstrating that noise correlation length — a measure of spatial autocorrelation in metabolic fluctuation patterns — reliably distinguishes neuroprotective from neurovulnerable states across phases of HIV infection. Provides a computationally tractable biomarker framework for neurometabolic monitoring in people living with HIV, with implications for early detection of HIV-associated neurocognitive disorders. The current version incorporates the first publicly available open-source consolidated dataset of phase-specific HIV neuro-metabolic MRS data — 49 observations across 7 independent human cohorts spanning four decades of published literature (1985–2021) — released as an independent scientific contribution to the HAND neuroimaging field.
A population-specific analysis of the structural, legal, and healthcare system barriers preventing people who inject drugs (PWID) from initiating long-acting injectable PrEP. PWID represent the highest-risk and lowest-coverage population in HIV prevention, with only 2% currently using PrEP despite indicated need. This work quantifies the compounding effect of criminalization, housing instability, stigma, and healthcare avoidance on bridge period completion, and evaluates syringe service program integration as the primary delivery pathway. Featured preprint in Preprints.org’s 10th anniversary edition (2026). Preprint downloads: 134 (Preprints.org, as of April 2026).
A formal derivation showing that the Kassanjee/Gao cross-sectional incidence estimator — the analytic backbone of counterfactual-controlled PrEP efficacy trials including PURPOSE 1 (NCT04994509) and PURPOSE 2 (NCT04925752) — produces systematically biased point estimates when applied to populations with structural censoring (overdose mortality, incarceration, displacement, carceral healthcare disruption). The paper derives the survival-biased effective MDRI Ω*(γ) and a joint IRR bias factor incorporating both screening-cohort and intervention-arm observation probabilities, then identifies the standard 90-day no-prior-testing eligibility criterion as a selection mechanism operating on the same testing-engagement axis that drives the competing-risk hazard. Applied to 34 high-burden US MSAs using AIDSVu 2023 surveillance data with late-diagnosis percentage as the empirical structural-hazard proxy: Kassanjee denominators inflated 8.7–27.3% with monotone scaling, BIRR ∈ [0.969, 0.999], reported IRRs attenuated up to 3.1% in the highest-severity empirical cohort — making interventions appear artificially superior in the populations trials ostensibly aim to serve.
A formal mathematical framework establishing that HIV post-exposure prophylaxis efficacy is governed by a finite, quantifiable prevention window determined by viral replication kinetics and the irreversibility of systemic infection establishment. Derives closed-form theorems for optimal PrEP timing under stochastic exposure conditions, with direct implications for same-day PrEP initiation protocols, bridge period design, and the pharmacokinetic rationale for conservative LAI-PrEP initiation requirements.
HIV-associated neurocognitive disorder (HAND) presents a cognitive paradox: patients maintain cognitive function during acute infection despite severe neuroinflammatory cytokine responses. Computational modeling demonstrates that quantum coherence is preserved substantially longer within microtubules arranged in Fibonacci geometries compared to regular structures, with strong correlations to neuroimaging biomarkers (r = 0.74–0.82, p < 0.001). The framework proposes a geometric-quantum mechanism for the HAND cognitive paradox. Full journal submission is held pending completion of an extension addressing the standard physics objection to quantum-biological coherence at physiological temperatures.
A cross-sectional audit of 9 major US medical journals finds that 0 of 14 editors-in-chief hold formal ML/AI training, first-author ML publications, or quantitative doctoral degrees. These journals collectively publish 0.115% of global PubMed-indexed ML/AI biomedical output (263 of 227,731 papers, 2023–2025). Predictive models estimate triage error rates of 45–78% for quantitatively intensive manuscripts submitted to these venues.
Open Science
All research code is open-source. All datasets are publicly archived. All tools are free for healthcare providers and researchers.
All computational tools and analysis code available at github.com/Nyx-Dynamics. GPG-signed commits. Pharma-Restricted Open Healthcare License.
Datasets and validation results archived at Zenodo DOI: 10.5281/zenodo.19363153 with permanent DOI assignment.
11 reviews completed (April 2025–present) spanning clinical ML for cardiovascular risk, infectious disease forecasting, multi-omics cardiology, NLP for vulnerable populations, and sports medicine ML. Median turnaround under 10 days. Two R1 revision rounds — handling editors route revisions back to the same reviewer (trust signal).
Dr. Demidont was formerly employed by Gilead Sciences (ending October 2024); all holdings divested December 2024. No industry funding. Disclosed per ICMJE guidelines in all publications.
Manuscripts developed with assistance from Anthropic Claude and other AI tools per ICMJE-compliant disclosure. Author maintains full responsibility for all scientific content and conclusions.