Approach
Science-driven methods for complex biomedical data
Biomedical data is becoming richer and more complex. More measurements can provide more information about diseases and optimal interventions to benefit patients, but they also create new statistical challenges.
Saddle Point Science develops mathematical, statistical and computational methods specifically designed to address these challenges and extract reliable insights from complex biomedical data.
Why a different approach?
Modern biomedical data creates new statistical challenges
Biomedical datasets often contain hundreds or thousands of measurements for a relatively small number of patients. The combination of such high-dimensional yet limited sample size data, with the complexities of measurements, leads to a varied range of analytical complexities, including:
-
Overfitting and estimation bias leading to poor reproducibility
-
Latent cohort heterogeneity and distinct patient subgroups
-
Informatively missing data (i.e. missing-not-at-random)
-
Complex, often competing clinical outcomes that cause informative censoring
-
Data distributed across multiple sites and organisations that cannot be easily shared
Traditional statistical approaches are not always designed for these conditions. Our approach is to develop methodology that explicitly accounts for the varied characteristics of modern biomedical data.
From scientific theory to practical solutions
Research-driven methodology, designed for real-world applications
Our work connects fundamental scientific research with practical biomedical applications.
Our Scientific Foundation
Mathematics, Statistics and Computational science
Our methodology draws on several complementary areas of scientific expertise.
-
Mathematical modelling: We use advanced mathematical theory to understand complex relationships within biomedical data.
-
Statistical inference: We develop methods for making reliable predictions and inferences from available data.
-
Bayesian methods: We use Bayesian approaches to quantify uncertainty and incorporate statistical evidence systematically into complex analyses.
-
Statistical mechanics: Concepts from statistical mechanics provide a foundation for addressing high-dimensional statistical problems and understanding overfitting.
-
Computational methods: We translate mathematical and statistical methodology into practical computational pipelines.
Designed for challenging datasets
Make better use of the data you already have
In biomedical research, collecting data for more patients is not always possible. This is particularly relevant when:
-
Diseases are rare
-
Clinical studies are expensive
-
Patient recruitment is difficult
-
Smaller data-sets are distributed across multiple centres but cannot be shared
These practical issues are compounded at the analysis stage when there are many measurements available per patient. Our methodology is designed to address these situations and make efficient use of all available information. The goal is not simply to analyse more data — but to extract more reliable information from the data available.
Our Solutions
Turning statistical methodology into practical tools
Addressing Overfitting
Reliable analysis when the number of variables is large
When the number of measured variables becomes large relative to the number of patients, conventional statistical models can become vulnerable to overfitting. Our methodology explicitly addresses this challenge through approaches including:
-
Probabilistic variable selection
-
Bias correction based on Replica method
-
Statistical regularisation
-
Analytical treatment of model parameters
-
Uncertainty assessment
These methods are incorporated into our software solutions to help generate more reliable predictions and statistical conclusions.
Reliability & Validation
We don't stop at generating a result
A useful statistical method should provide ways to assess how reliable its results are. Our analytical solutions incorporate validation and quality-control approaches such as:
-
External validation: Test predictions using additional datasets where available.
-
Simulated data: Assess model behaviour using controlled simulated datasets.
-
Shadow analysis: Use outcome-randomised data to investigate potential false-positive findings.
-
Uncertainty assessment: Quantify uncertainty associated with predictions and subgroup membership.
-
Reproducible workflows: Use structured and automated analytical pipelines to support reproducibility.
-
Automated reporting: Generate reports documenting analytical results and key findings.
From Methodology to Software
Making advanced statistics practical
Advanced statistical methodology should not remain limited to theoretical research. We translate our research into automated software solutions that allow researchers and organisations to apply sophisticated statistical approaches to real-world biomedical datasets.
Our solutions are designed to:
-
Automate complex analytical workflows
-
Reduce manual processing
-
Support reproducible analysis
-
Generate structured reports
-
Make advanced methodology practical for research applications
Our approach in practice
Turning statistical methodology into practical tools
spsSIGNATURE
Personalised risk and treatment response predictions
Generate reliable individual-level predictions from complex, high-dimensional biomedical data.
Research-driven, Application-focused.
From scientific innovation to biomedical impact
Our approach brings together scientific theory and advanced statistical methodology to create advanced-yet-practical analysis pipelines. We continuously develop and validate our methods through scientific research and collaboration with academic, healthcare and industry partners.
Do you have a challenging dataset?
Let's explore what your data can reveal.
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.