SAXS + chromatography
Resolve mixture states, generate structural signatures and validate inferred states. Pair-distance distributions, peak deconvolution and time-resolved signals become structured inputs.
Projects
One core idea, multiple instrument and analysis domains. Calyr.ai is built around a single representation layer that converts experimental signals into a parameter space where physical models and prediction can connect.
Core concept
Parameterized PCA as a universal representation of experimental signals
Experimental observables are high-dimensional. PCA extracts modes. Parameterization turns modes into interpretable coordinates that can be coupled to physics models, simulations and learning systems.
Goal: map experimental observables → parameter space → predictive models.
Method layer
A reusable pipeline that stays stable across domains.
Experiments
SAXS, chromatography, cryo-EM, trajectories, fluxes
Signal representation
preprocessing, normalization, aligned observables
PCA modes
low-dimensional modes of variation
Parameterized space
interpretable coordinates linked to physical parameters
Physics / ML models
MD, thermodynamics, flux models, learned predictors
Prediction layer
hypotheses, parameter inference, experiment design
Experimental platforms
Different measurement domains become data sources and validation layers for the same representation.
Cryo-EM
Structural validation and state classification. Connect inferred modes to real conformational ensembles via density maps and discrete classes.
Molecular dynamics
Continuous trajectories and energy landscapes. Relate PCA coordinates to free-energy surfaces and mechanistic intermediates.
Metabolic modelling
Systems-level extension. Replace structures with flux states and thermodynamic constraints, but keep the same modal representation.
AI / prediction layer
LLM-assisted modelling is treated as an interface layer: it translates experimental descriptions and parameter constraints into model setups, simulation prompts and hypothesis candidates.
Experiment → model suggestion
Parameter inference and uncertainty framing
Automated hypothesis generation and validation loops
Integration layer
The message: the same analytical structure works across domains. A unified representation (PCA coordinates + parameterization) can be applied to scattering signals, chromatographic pulses, trajectories and fluxes.
Subprojects
Concrete implementations and prototypes around the core concept.
Public
Private / internal
Roadmap, experiments and internal work.