spsSIGNATURE
Reliable predictions from complex biomedical datas
spsSIGNATURE helps researchers predict clinical outcomes and identify patient groups based on individual risk and treatment response. It is designed to address two common challenges in biomedical research: overfitting and statistical bias, which can make results less reliable, especially when datasets contain many variables but few patients.
The software supports different types of clinical outcomes, including survival and time-to-event data, ordinal class outcomes and continuous clinical measurements. It is designed to work with both small datasets and high-dimensional data, helping researchers make better use of the information available. By addressing overfitting and statistical bias, spsSIGNATURE supports more reliable predictions and patient stratification, even when large datasets are difficult to obtain, such as in rare disease research or smaller clinical trials. It can also correct for informative missingness of covariates and accommodate additional variables, including interactions between risk factors, to help researchers explore how different factors may work together.
A different approach to overfitting
Unlike conventional regularisation methods, spsSIGNATURE uses statistical methods based on the replica method, a mathematical framework for analysing complex statistical models. This approach allows the software to address overfitting without setting aside data for hyperparameter optimisation. In bias-removal mode, hyperparameters are not required; in regularisation mode, they are calculated analytically. Moreover, unlike standard regularization strategies, it is able to correct also all auxiliary model parameters. The result is a streamlined analytical process designed to make efficient use of available data while supporting reliable statistical conclusions.
What can spsSIGNATURE help you achieve?
-
Risk prediction
-
Generate personalised estimates of clinical risk.
-
-
Treatment response prediction
-
Identify patterns associated with differences in treatment response.
-
-
Patient stratification
-
Identify meaningful differences between patients based on their predicted outcomes.
-
-
Personalised treatment
-
Estimate personalised optimal drug doses where applicable.
-
Designed to reduce overfitting
High-dimensional biomedical datasets can contain many more variables than available samples, making reliable prediction challenging.
spsSIGNATURE addresses this overfitting regime through a combination of automated variable selection, mathematical bias correction and statistically derived regularisation. This allows the pipeline to make efficient use of available data without sacrificing data for conventional hyperparameter optimisation.
Advanced statistical methodology
spsSIGNATURE is based on statistical mechanical theory developed by Saddle Point Science in collaboration with academic partners. The methodology uses the replica method to derive analytical corrections for overfitting in Generalized Linear Models (GLMs).
Key methodological features
-
Probabilistic variable selection
-
Bias correction
-
Analytical regularisation
-
Correction for nuisance parameters
-
Handling of informative covariate missingness
-
No data sacrifice for hyperparameter optimisation
Built-in validation and quality control
spsSIGNATURE includes tools designed to assess the reliability of predictions and reduce the risk of false-positive findings.
-
Shadow analysis : Analyse outcome-randomised data to identify potential false-positive inferences.
-
External validation: Validate predictions on additional datasets.
-
Simulated clinical data: Generate simulated datasets to assess model behaviour.
-
Automated reporting: Generate comprehensive analysis reports automatically.
Publications in which the pipeline was used
-
Barber, P. R., Mustapha, R., Flores-Borja, F., Alfano, G., Ng, K., Weitsman, G., Dolcetti, L., Mohamed, A., Wong, F., Vicencio, J. M., Galazi, M., Opzoomer, J. W., Arnold, J. N., Thavaraj, S., Kordasti, S., Doyle, J., Greenberg, J., Dillon, M. T., Harrington, K. J., . . . Ng, T. (2022). Predicting progression-free survival after systemic therapy in advanced head and neck cancer: Bayesian regression and model development. eLife, 11. https://doi.org/10.7554/elife.73288
-
Colijn, J. M., Meester-Smoor, M. A., Verzijden, T., De Breuk, A., Silva, R., Merle, B. M. J., Cougnard‐Grégoire, A., Hoyng, C. B., Fauser, S., Coolen, A. C. C., Creuzot‐Garcher, C., Hense, H., Ueffing, M., Delcourt, C., Hollander, A. I. D., & Klaver, C. C. W. (2021). Genetic Risk, Lifestyle, and Age-Related Macular Degeneration in Europe. Ophthalmology, 128(7), 1039–1049. https://doi.org/10.1016/j.ophtha.2020.11.024
-
Grigoriadis, A., Gazinska, P., Pai, T., Irhsad, S., Wu, Y., Millis, R. R., Naidoo, K., Owen, J., Gillett, C., Tutt, A., Coolen, A. C. C., & Pinder, S. (2018). Histological scoring of immune and stromal features in breast and axillary lymph nodes is prognostic for distant metastasis in lymph node‐positive breast cancers. The Journal Of Pathology, 4(1), 39–54. https://doi.org/10.1002/cjp2.87
Are you ready to explore what your data can reveal?
Do you have a challenging biomedical dataset or research question?
Whether you are working with a small dataset, high-dimensional measurements, heterogeneous patient populations or distributed clinical data, our team can help explore whether our approach is suitable for your research question.