Services

Predictive Patterns offers a wide variety of software development and data analysis services. In particular:

Rapid Application Development

We are experienced in the full software development life-cycle: requirements gathering, specification, development, testing, user documentation, deployment, support and maintenance. We make maximum use of existing open-source libraries and tools, as well as using state-of-the-art XML-based code-generation technology to handle much of the rote work that can take time and resources away from implementing the core logic of your application.

Hardware Integration

Leading edge hardware requires leading edge software to go with it. Whether it's a USB device driver or an application front-end, we can ensure that your customers get the best user experience possible, and that the science your hardware emboidies is presented to them in a comprehensible, well-documented way.

Image Processing

We have extensive experience in image processing algorithms. Predicitive Patterns' founder, Tom Radcliffe, invented the pseudo-correlation image registration algorithm in 1991, which was the first algorithm to use sampled pixel data as the basis for registration, resulting in enormous speed gains over the conventional cross-correlation approach. Other algorithms, such as mutual information, have subsequently adopted a similar technique. We are experts in the use of the VTK's image processing system as a platform for developing new image processing algorithms, and have most recently worked in segmentation and pattern recognition.

Data Analysis

We specialize in genomics data analysis, working closely with university researchers on a number of projects in cancer genomics. Our founder, Tom Radcliffe, is an Adjunct Assistant Professor in the Department of Pathology and Molecular Medicine at Queen's University. We use both traditional statistical methods (F-test, Kruskal-Wallis, etc.) and advanced pattern discovery algorithms to mine whole-genome datasets for significant patterns of expression. The most recent addition to our proprietary algorithmic toolkit is capable of finding patterns in a few dozen genes out of 41,000 that robustly distinguish sample classes, even in cases when no single gene or gene pair shows better than random differentiation. Despite our use of leading-edge algorithms, we think what Kronecker actually meant was, "God made the p-values, all else is the work of man." We take a conservative, objective approach to statistics, and use a variety of Monte Carlo and formal methods to ensure that no result lacks an objective measure of statistical significance. Although our focus has been on genomics in the past, we are aware that many of the same algorithms apply equally well to other kinds of data, from finance to engineering.

Everything Else

We are eclectic. While the above areas list our specialties, our experience is not limited to them. On staff we have a published poet and an aspiring novelist, as well as people with more predictable expertise in C++, Java, Perl, Python... even FORTRAN. We have done work in fluid modelling and Stirling engine design, for which we developed an adaptive C++ implementation of a 4th order Runge-Kutta ODE solver, and we have spent more time doing Monte Carlo simulations of radiation and particle transport than we care to think about. So if you find yourself with a problem that is amorphous and ill-defined, or that seems to take a peculiar combination of skills, give us a call!