AIVC platform

Biological intelligence starts
with the right data.

We are building a proteomics-rich approach to cellular prediction: generate informative experiments, learn biological responses, and test the next decision.

Our 3+1 framework

Three data pillars.
One learning loop.

Cell Research · 2025

A priori knowledge

Pathways, molecular interactions, literature and mechanistic priors.

Static architecture

Cellular and tissue organization, spatial proteins, DNA and images.

Dynamic states

Interventions, dose and time, molecular responses and independent functional outcomes.

Closed-loop active learning

Predict a response. Identify uncertainty. Select informative experiments. Measure the outcome and refine the model.

Framework proposed by Liujia Qian, Zhen Dong and Tiannan Guo, Cell Research (2025). The learning loop guides data generation; it is not a fourth data modality.

Protein-rich, multimodal by design

Different layers reveal
different aspects of a cell.

Our focus is purpose-designed intracellular proteomics linked to perturbation, time and spatial context. Other data layers contribute complementary evidence.

Data layer	What it contributes	Why complementary data matter
scRNA-seq / Perturb-seq	Cell populations and transcriptional responses.	Protein turnover, modifications and activity are not directly measured by RNA counts.
Imaging	Morphology, organization, selected protein markers and live dynamics.	Molecular depth depends on markers, resolution and acquisition.
Clinical records	Treatment histories, disease trajectories and patient outcomes.	Observational evidence differs from controlled cellular intervention experiments.
Affinity proteomics	Sensitive profiling of predefined targets; widely used in plasma and serum.	Binding reagents define the target menu. Validated tissue and cell applications also exist.
WO spatiotemporal MS strategy	Broad intracellular protein states, controlled perturbations and spatial sampling.	Connect direct measurements with functional assays; test the added value of joint time–space modeling.

Measurement coverage and throughput depend on the workflow. Protein abundance does not, by itself, establish activity or causal mechanism.

Technical foundations

Four connected research capabilities.

Spectral intelligence

MassNet · DIA-BERT · DDA-BERT

Turn mass-spectrometry spectra into interpretable protein evidence with AI-supported analysis.

Explore spectral AI

Perturbation dynamics

ProteinTalks

Learn how protein states change after intervention, and evaluate drug-response and combination tasks.

Read ProteinTalks

Spatial context

FAXP

Connect protein measurements to tissue location, cells and organelles through image-guided sampling.

Read the spatial method

Multimodal integration

Knowledge · architecture · dynamics

Combine complementary molecular layers with biological priors and function-centered model design.

Read the data framework

From foundations to an integrated model

ProteinTalks supplies published temporal-response evidence. FAXP supplies spatial measurement methods. Our integrated strategy connects these components through matched experiments and independent model evaluation.

Research partnerships

Build around a defined decision.

Custom perturbation datasets

Agree the biological context, interventions, controls and endpoints. Generate standardized profiles with QC and metadata.

Model co-development

Develop a vertical predictive model with held-out tests and independent experimental confirmation.

Deployment partnerships

Explore qualified local deployment and integration for an appropriate partner environment.