Prediction in biology: populations, error estimates, models and mathematics

Mike Bittner
Plenary Talk, Morning Session, 2 September (11th MGED Meeting, 1-4 September, 2008)

How close are we to getting into the clinic? Over the past few years, deaths by cardiovascular disease has dropped dramatically, while cancer has remained about the same. In fact, there are now more deaths from cancer than from cardiovascular in the States - since the late 1990s. Molecular biology was begun long ago with the help of the influx of a large number of nuclear physicists.

What are molecular biologists missing? Some blame our rapid  and changing use of technology, which causes a form of amnesia about what went on before.

Confounding assumptions in the evaluation of data for model building:
independence assumption. Optimal number of features depends on sample size, classification rule and feature-label distribution. It's good to use LDA, linear model, and highly-correlated features.

[a large part of the talk is missing here due to problems with my laptop that required a restart. My apologies!]

Canalyzing genes force all effort down one (or a small proportion) of the possible pathways.

These are just my notes and are not guaranteed to be correct. Please feel free to let me know about any errors, which are all my fault and not the fault of the speaker. :)