Fragmented variables
- Heterogeneous names, units and intervals
- Domain-specific isolated endpoints
- Manual integration that is difficult to reproduce
- ML fed directly with raw variables
AXIS Research
Research generates variables - AXIS builds comparable states.
Deterministic infrastructure for turning blood, fecal microbiota, metabolomics, genetics, phenotypes, interventions and time into reproducible participant and cohort representations. Within-domain patterns and cross-patterns can be frozen before statistics or machine learning are applied.
01 / Scientific bottleneck
The problem is not always obtaining more data - It is making them comparable.
Multimodal studies accumulate source families with different scales, frequencies and meanings. AXIS preserves provenance and creates a prior structure layer for within-domain patterns and cross-patterns without turning research into a black box.
Reproducible states
- Canonical identity and provenance preserved
- Within-domain patterns and cross-patterns defined by protocol
- Confidence, coverage and conflicts explicit
- Structured features before statistics or ML
02 / Research architecture
One frozen configuration. Comparable states across every participant and visit.
AXIS converts heterogeneous study data into a version-consistent state matrix. The protocol, variables and inference rules are fixed before execution, so every result can be reproduced and traced back to its original measurement.
Freeze the scientific contract
The question, population, domains, interventions, visits and permitted outputs are mapped before the first execution.
- Protocol mapDefines who is studied, what is measured and when.
- Canonical dictionaryFixes variable identity, units, provenance and availability.
Apply one explicit inference path
Accepted inputs are normalized and processed only through the parameters and rules defined for that study.
- Research parametersCanonical signals interpreted under study-specific SIRs.
- Patterns → blocks → axesExplicit relationships are executed while conflicts remain visible.
Build comparable evidence
Each participant and visit becomes a structured state vector before alignment at cohort level.
- Subject stateAxes and, when specified, DBSI — with confidence, coverage and coherence.
- Cohort matrixParticipants × visits, subgroups, trajectories and governed exports.
03 / Illustrative cohort state
One shared representation across participants and time.
The matrix is conceptual and does not represent scientific results. It shows how a study can compare structured dimensions without losing participant-level evidence paths.
- Comparable state vectors at every visit
- Explainable subgroup descriptions
- Pattern prevalence and coverage analysis
- Longitudinal trajectories
- Exports for statistics and machine learning
Cohort stateN = 005 · T = 003 · research output
Subject
MET
INF
O₂
REN
COH
P-001
62
69
76
83
90
P-002
73
80
87
94
67
P-003
84
91
64
71
78
P-004
95
68
75
82
89
P-005
72
79
86
93
66
04 / Scientific outputs
Structure before inference.
AXIS provides deterministic features and evidence paths. Statistical conclusions remain a subsequent methodological layer.
✓
Participant state vectors
Structured dimensions with confidence, coverage and provenance.
✓
Cohort matrices
Version-consistent representations for comparison and validation.
✓
Explainable subgroups
Describe groups through activated patterns, not only mathematical distance.
✓
Structured trajectories
Retain what changed, with what confidence and which evidence supported the transition.
✓
Pattern prevalence
Estimate how often defined relationships occur in a cohort.
✓
ML-ready exports
Stable, interpretable features for downstream modeling.
Research boundary
Research configurations do not automatically inherit clinical validity, utility or commercial claims. Lifecycle state, configuration version and evidence status must remain explicit.
Research collaboration
Design a reproducible cohort around a defined biological question.
The strongest collaborations begin with the hypothesis, data contract, cohort definition and validation plan—not with an unrestricted parameter upload.