Saddle Point Science was created with the aim of raising the standards of mathematical and computational tools commonly applied in the biomedical sciences. Since the human genome project, biomedicine presents us with clinical and genomic data of unprecedented complexity and volume. Yet, most mathematical and statistical methods commonly used to analyse such data date from the 1970s. As a result, many expensive medical data sets are significantly under-utilised, therapeutic claims in clinical studies are often found to be non-reproducible, and targeting of expensive drugs and treatments with serious side-effects is of limited precision.
In recent decades there has been too much focus in medical data analysis on uniformity of statistical and modelling methods, at the expense of innovation. It should not be common practice to limit one's analysis of clinical data to the calculation of hazard ratios and p-values, just because these are coded in off-the-shelf software. Uncritical use of simple tools in complex situations may cause important signals to be missed, and can lead to erroneous conclusions - thereby causing patient harm. We believe that progress in medical data analysis requires the integration of Bayesian statistical and information-theoretic ideas with advanced modelling techniques from theoretical physics.
We develop advanced mathematical and statistical techniques, modelling approaches, and efficient software tools, for the analysis of complex epidemiological and medical trial data. Our emphasis is on confronting what we see as the main challenges of modern medical data: dimension mismatch (too many covariates, too few samples), latent heterogeneity in disease or host (not visible in the covariates), informative censoring, and confounding factors. We also develop mathematical tools for modelling and manipulating cellular and immunological signalling processes.