Helix: Genome-Guided Personalization of Human Enhancement Compounds

Authors: Sowmay Jain¹, Lauki Antonson² (AI Research Agent)
Affiliations: ¹Upsurge Labs (upsurge.io), ²Autonomous AI Agent (lauki.ai)
Date: April 2026

1. Abstract

We propose a genome-guided framework for predicting individual response to human enhancement compounds across three domains: muscle performance, cognitive function, and mood regulation. Using whole-genome sequencing (Oxford Nanopore, 30x coverage) and a validated bioinformatics pipeline (minimap2 for alignment, Clair3 for variant calling, SnpEff/SnpSift for annotation against ClinVar, PharmGKB, and gnomAD), we have completed a full N=1 proof of concept. This pilot extracted 141 drug-gene interactions across 62 pharmacogenes (CPIC Level A/B evidence), computed polygenic risk scores for major depressive disorder using PGC MDD2025 summary statistics (267 independent loci), and mapped enhancement-relevant variants across metabolic, neurotransmitter, and musculoskeletal pathways.

Our team consists of Sowmay Jain (human subject and principal investigator) and Lauki Antonson, an autonomous AI research agent that independently built and executed the entire genomic pipeline — from downloading 67GB of raw FASTQ data to producing clinically annotated enhancement profiles. All analysis runs on local infrastructure with zero external data transfer, ensuring complete genomic privacy. With seed funding, we will scale to a 5-subject pilot cohort, open-source the full pipeline, and build a public genome-to-enhancement profiling dashboard. Estimated timeline: 6 months.

2. Introduction

Human enhancement research overwhelmingly treats subjects as a homogeneous population. Creatine trials report average effect sizes across all participants. Modafinil studies aggregate responders and non-responders into a single mean. SSRI efficacy is measured as a population-level statistic. Yet decades of pharmacogenomics research have demonstrated that individual genetic variation causes dramatic, clinically meaningful differences in compound metabolism, receptor sensitivity, and downstream signaling cascades.

The gap is clear: no systematic framework currently exists for mapping whole-genome sequence (WGS) data to predicted response across enhancement compounds spanning muscle performance, cognitive function, and mood regulation. Clinical pharmacogenomics focuses narrowly on drug safety — identifying patients who will experience adverse reactions. The extension of this same framework to human optimization — predicting which compounds will be most effective for a given individual — remains largely unexplored in the literature.

We have already built this system. Not as a theoretical proposal, but as a working pipeline validated on real human DNA. Using Oxford Nanopore whole-genome sequencing at 30x coverage (5.8 million called variants), we have:

Annotated all variants against ClinVar, PharmGKB, gnomAD, and dbSNP using SnpEff/SnpSift
Extracted 141 drug-gene interactions across 62 pharmacogenes at CPIC Level A/B evidence
Profiled 27 longevity-associated SNPs across 40 genes (FOXO3, APOE, mTOR pathway, CETP, KLOTHO)
Screened mood-relevant loci across the HPA axis (FKBP5, NR3C1, CRHR1), serotonergic (SLC6A4, HTR2A, TPH2), dopaminergic (COMT, DRD2, ANKK1), and neurotrophic (BDNF) pathways — 26,518 variants extracted from mental health-associated genes
Computed polygenic risk scores for major depressive disorder (PGC MDD2025, 267 independent loci) — subject scored at the 7.6th percentile (z = −1.43, very low genetic risk)
Mapped nutrient metabolism variants affecting Vitamin D, B12, iron, calcium, and folate absorption

When we shared this work publicly (x.com/sowmay_jain/status/2041982135305957425), the response was substantial — researchers, biohackers, and clinicians all asking: how do I replicate this for my own genome? The demand is demonstrated. What remains is making it accessible: a public dashboard, open methodology, and reproducible pipeline that anyone with WGS data can use.

We propose to use this funding to: (1) build a public genome-to-enhancement profiling dashboard, (2) open-source the complete pipeline so others can replicate it, (3) scale to a 50-subject cohort to validate genotype-response mappings, and (4) publish all results as an open-access resource on ResearchHub.

3. Hypotheses

Primary hypothesis (H1): Individual genomic variation, identifiable through whole-genome sequencing, predicts differential response to human enhancement compounds across three domains — muscle performance (creatine, beta-alanine, caffeine for endurance), cognitive function (modafinil, racetams, caffeine, choline precursors), and mood regulation (SAMe, methylfolate, adaptogens, SSRI sensitivity).

H2 (Cognitive stratification): Pharmacogene metabolizer status (CYP1A2, CYP2D6, CYP2C19, COMT) stratifies cognitive enhancer responders vs. non-responders with clinically meaningful effect size differences (Cohen's d > 0.5 between genotype groups).

H3 (Muscle performance): Muscle performance gene variants (ACTN3 R577X, ACE I/D, PPARGC1A Gly482Ser, MSTN) predict differential response to strength vs. endurance enhancement protocols.

H4 (Mood compound efficacy): HPA axis and serotonergic pathway variants (FKBP5, 5-HTTLPR, MTHFR C677T, COMT Val158Met) predict mood compound efficacy. Specifically, MTHFR 677 TT carriers are expected to show superior response to methylfolate vs. folic acid, and COMT Val/Val carriers are expected to show differential response to SAMe supplementation.

4. Methods

4.1 Study Design

This is a prospective observational cohort study with pre-registration of all genotype-response predictions. Phase 1 enrolls N=5 healthy adults (ages 20–45), stratified by sex. Phase 2 (contingent on crowdfunding) scales to N=50. Each subject undergoes whole-genome sequencing (Oxford Nanopore, minimum 20x coverage) and provides baseline assessments across all three enhancement domains before beginning compound protocols.

4.2 Genomic Pipeline (Validated — Already Operational)

The complete bioinformatics pipeline has been validated on our N=1 subject with the following performance metrics:

Alignment: minimap2 (Nanopore long-read mode) — 99.83% mapping rate achieved on N=1
Variant calling: Clair3 (Nanopore-optimized deep learning caller) — 5.8 million variants called
Annotation: SnpEff + ClinVar + PharmGKB + gnomAD — 149MB annotated VCF produced
Pharmacogenomic extraction: Custom pipeline across 62 pharmacogenes — 141 drug-gene interactions mapped
Polygenic risk scores: PGC GWAS summary statistics — MDD and SCZ scores computed
Enhancement profiling: Novel framework (to be formalized in this study)

All software components are open-source. The pipeline runs entirely on local infrastructure — no genomic data is transmitted to external services at any point.

4.3 Enhancement Compound Panels

Muscle Panel: Creatine monohydrate (5g/day, 8 weeks) stratified by ACTN3 rs1815739, ACE rs4646994, and CKM rs8111989. Beta-alanine (3.2g/day, 8 weeks) stratified by CNDP1 D18S880 repeat and histidine metabolism variants. Caffeine for endurance stratified by CYP1A2 rs762551 metabolizer status.

Cognition Panel: Caffeine (200mg acute dose) stratified by CYP1A2 metabolizer status and ADORA2A rs5751876 anxiety sensitivity. L-theanine + caffeine (200mg/100mg) stratified by COMT Val158Met rs4680. Alpha-GPC (600mg/day, 4 weeks) stratified by CHRNA7 rs6494223 and choline transporter variants. Creatine for cognition (5g/day, 6 weeks) stratified by GAMT, GATM, and SLC6A8 variants.

Mood Panel: Methylfolate (15mg/day, 8 weeks) stratified by MTHFR rs1801133 (C677T) and rs1801131 (A1298C). SAMe (400mg/day, 6 weeks) stratified by COMT Val158Met and MTHFR status. Omega-3 (2g EPA/day, 12 weeks) stratified by FADS1/FADS2 rs174546 and rs174575 desaturase efficiency. Rhodiola rosea (400mg/day, 4 weeks) stratified by FKBP5 rs1360780 and NR3C1 BclI polymorphism for HPA axis sensitivity.

4.4 Outcome Measures

Muscle outcomes: 1RM bench press and squat (standardized protocol), time-to-exhaustion cycling ergometry, and DEXA body composition — measured at baseline, 4 weeks, and 8 weeks.

Cognitive outcomes: N-back working memory task (2-back, 3-back), psychomotor vigilance test (PVT) for reaction time, and Stroop/Trail Making B for executive function — measured at baseline, acute post-dose, and 4 weeks.

Mood outcomes: PHQ-9, GAD-7, and POMS self-report questionnaires at baseline, 2, 4, and 8 weeks. Salivary cortisol (AM/PM) via ELISA at baseline, 4, and 8 weeks. Inflammatory markers (hs-CRP, IL-6) from blood draws at baseline and 8 weeks.

4.5 Control Variables

Sleep is tracked via wrist-worn actigraphy plus sleep diary throughout the study period. Diet is recorded via 3-day food records at baseline and weeks 4 and 8. Exercise follows a standardized training protocol for the muscle panel; habitual activity is logged for cognition and mood panels. A 2-week washout period separates compound phases, with 48-hour caffeine restriction before acute cognitive testing sessions.

5. Statistical Analysis Plan

For each compound-genotype combination, we will test whether genotype group predicts outcome using linear mixed models with compound response (change from baseline) as the dependent variable, genotype group as the fixed effect, and subject as a random effect for crossover compounds. Genotype-by-compound interaction terms will test whether the genotype effect differs meaningfully from zero. All primary genotype-compound predictions will be pre-registered with directional hypotheses (e.g., "CYP1A2 slow metabolizers will show greater anxiety increase from 200mg caffeine than fast metabolizers").

Effect sizes will be reported as Cohen's d for each genotype stratum with 95% confidence intervals. Multiple comparison correction uses Bonferroni within each domain (approximately 9 tests for muscle, 12 for cognition, 12 for mood) with FDR (Benjamini-Hochberg) as a secondary threshold. Pre-registration of all primary predictions eliminates the garden-of-forking-paths problem.

Polygenic risk score analysis: We will compute PRS for relevant traits (lean body mass, educational attainment, MDD, neuroticism) using PGC and UK Biobank summary statistics, testing PRS as a continuous predictor of compound response and reporting R-squared increment from adding PRS to single-gene predictor models.

Power: With N=50 and genotype frequencies of 20–50% for key variants (CYP1A2 slow metabolizer: ~46%; COMT Val/Val: ~25%; MTHFR 677 TT: ~10–12%), we have 80% power to detect Cohen's d of 0.7 or greater between genotype groups at alpha = 0.05. This threshold is appropriate for pharmacogenomic effects, which typically show large effect sizes (d = 0.5–1.5 for metabolizer status).

6. Ethics and Data Management

All genomic analysis runs on local infrastructure with zero external data transfer. Individual-level genomic data will be available only through controlled access mechanisms to protect participant privacy. Summary-level results, all analysis code, and the complete pipeline will be published as open-access resources. Negative results will be reported with equal emphasis — determining that a genotype does not predict compound response is as scientifically valuable as confirming that it does.

7. Pilot Data

We have completed a full N=1 proof of concept on subject SO_13000_SJ (Oxford Nanopore WGS, 30x coverage). Key findings with direct enhancement implications:

CYP1A2 rs762551 (*1F): C/A heterozygous — slow caffeine metabolizer. Enhancement implication: cap caffeine at approximately 200mg, avoid afternoon dosing for cognitive enhancement protocols.
COMT rs4680 (Val158Met): G/G Val/Val — "Warrior" genotype with fast dopamine clearance, associated with stress resilience and higher pain tolerance. May respond well to dopamine-sparing cognitive compounds.
CYP2C19 rs4244285 (*2): A/A homozygous — poor metabolizer. SSRIs (citalopram, escitalopram, sertraline) will accumulate to potentially toxic levels. Dose reduction mandatory. This is a critical pharmacogenomic safety finding.
MDD PRS (PGC 2025, 267 loci): z = −1.43, 7.6th percentile — very low genetic depression risk. Mood fluctuations are likely situational rather than endogenous, informing compound selection toward adaptogens over serotonergic interventions.
APOE: epsilon-3/epsilon-3 — neutral Alzheimer's/CVD risk, providing a green light for aggressive cognitive enhancement protocols.
FOXO3 rs2802292: A/A reference — missing the longevity-associated allele, suggesting prioritization of autophagy-activating interventions (fasting protocols, exercise, rapamycin research).

This single genome produced actionable enhancement guidance across all three target domains in under 48 hours of compute time. Results are currently hosted at helix.lauki.ai (access-controlled due to raw genomic data). With funding, the pipeline, methodology, and anonymized results will be published as a fully open-access resource.

8. Budget

Phase 1 — Seed (10,000 USD): N=5 Pilot + Public Platform

Whole-genome sequencing (5 subjects at 400 USD each): 2,000 USD — Oxford Nanopore, outsourced to certified lab
Compute (alignment + variant calling): 200 USD — cloud GPU, approximately 2 hours per sample
Enhancement compounds (5 subjects, 3 domains): 500 USD — creatine, caffeine, methylfolate, SAMe, alpha-GPC
Cognitive testing software license: 500 USD — computerized assessment battery
Blood draws + cortisol ELISA (5 subjects, 2 timepoints): 750 USD
Public Helix dashboard development: 4,000 USD — open-source genome-to-enhancement profiling tool
Pipeline documentation and reproducibility package: 1,000 USD
Preprint preparation and publication: 1,050 USD

Phase 2 — Crowdfunding Stretch (40,000 USD): Full N=50 Cohort

Additional WGS (45 subjects at 400 USD each): 18,000 USD
DEXA scans (50 subjects, 2 timepoints): 5,000 USD
Full blood panels + ELISA (50 subjects, 3 timepoints): 7,500 USD
Actigraphy devices (50 units): 5,000 USD — wrist-worn sleep/activity trackers
Research assistant (part-time, 6 months): 4,500 USD — data collection and subject coordination

All genomic analysis software is open-source (minimap2, Clair3, SnpEff, bcftools). Complete pipeline code will be published alongside the study results.

9. References

Relling MV, Klein TE. CPIC: Clinical Pharmacogenetics Implementation Consortium of the Pharmacogenomics Research Network. Clin Pharmacol Ther. 2011;89(3):464–467.
Zheng Z, et al. Clair3: Accurate small variant calling for long reads. Nat Methods. 2022;19(10):1265–1272.
Li H. Minimap2: Pairwise alignment for nucleotide sequences. Bioinformatics. 2018;34(18):3094–3100.
Cingolani P, et al. A program for annotating and predicting the effects of single nucleotide polymorphisms, SnpEff. Fly. 2012;6(2):80–92.
Wray NR, et al. Genome-wide association analyses identify 44 risk variants and refine the genetic architecture of major depression. Nat Genet. 2018;50(5):668–681.
Guest NS, et al. International Society of Sports Nutrition Position Stand: Caffeine and exercise performance. Med Sci Sports Exerc. 2021;53(3):618–628.

Proof of Work: Full story of the N=1 genome analysis: x.com/sowmay_jain/status/2041982135305957425

Authorship: Sowmay Jain (principal investigator) and Lauki Antonson (AI research agent).
Contact: hi@lauki.ai | sowmay@upsurge.io
Organization: Upsurge Labs (upsurge.io)

Helix: Genome-Guided Personalization of Human Enhancement Compounds

Helix: Genome-Guided Personalization of Human Enhancement Compounds

Key Insights

Peer Reviews

Related Funding

Drug Discovery for human performance and function optimization

Topics