Prognostic Risk Integrative Stage Model

PRISM-GI

A pan-gastrointestinal cancer prognostic tool that combines a 15-gene expression-derived risk score with pathological stage to estimate 1-, 3-, and 5-year survival probability.

Start Contact

15 literature-guided genes

TCGA training cohort

GEO external validation

Risk score plus pathological stage nomogram preview — Risk score + pathological stage nomogram

Browser calculator

Calculate PRISM-GI risk and survival probabilities

Paste a cohort-level expression matrix. All computation runs locally in the browser; no patient data is uploaded.

Input guidance. TPM, FPKM, log-transformed normalized expression, normalized counts, and microarray-normalized expression can be used if all samples in the uploaded cohort use the same scale. Raw counts should be normalized for library size first. Ct values are not recommended unless transformed so larger values mean higher expression.

Cohort input

Required columns: SampleID, Stage, and the 15 PRISM-GI genes. Stage accepts I, II, III, IV, or 1-4.

Required genes

Results

Load example data or paste your cohort matrix.

Sample	Stage	Risk score	Risk group	1-year survival	3-year survival	5-year survival

Deployment note

The research model was trained using rank-expression features derived from a broader literature-mined gene universe. This web calculator uses the required 15 genes to compute within-sample percentile ranks, then applies cohort centering, the TCGA-trained random-subspace Cox model, and the TCGA-trained risk + stage Cox nomogram. Use it for cohort-level exploratory analysis and validation, not as a clinical diagnostic device.

Contact and lab

Weikaixin Kong

Postdoctoral Researcher, School of Computer Science, Aalto University.

Associate Professor, School of Biomedical Engineering and Technology, Tianjin Medical University.

Research focus: computational pathology, multi-omics modelling, cancer prognosis, interpretable machine learning, and translational biomedical AI.

CPM Lab

The lab develops computational models for biomedical data, with an emphasis on robust prediction, cross-cohort validation, and clinically interpretable biomarkers.

Visit cpmlab.cn