DOE Human Genome Program Contractor-Grantee
57. Automated Optimization of Expert System for Base-Calling in DNA Sequencing
Arthur W. Miller and Barry L. Karger
Barnett Institute, Northeastern University, 360 Huntington Ave., Boston, MA 02115
A recurring issue in automated DNA sequencing is that base-calling lags behind improvements to instrumentation and sequencing chemistry. This is because base-callers require retraining, or because the preprocessing of the data prior to base-calling must be changed. We have previously presented an expert system for long-read base-calling, capable of read lengths up to 1300 bases in sequencing by capillary electrophoresis on optimized separation matrices (A. W. Miller and B. L. Karger, DOE Human Genome Program Contractor-Grantee Workshop VII, 1999). The expert system supplies probabilistic confidences on base-calls, with statistics computed for several different types of miscall. Here we present tools for the automated retraining and optimization of this base-caller, including preprocessing and confidences, by nonprogrammers. Training takes into account template effects, low signal, and other factors observed in production sequencing. Results are shown for large amounts of data from both ABI 3700 and MegaBACE 1000 sequencers. In addition to software, other recent developments in long-read sequencing by capillary electrophoresis will also be presented.
This work is being supported by DOE grant DE-FG02-90ER 60985.
|The online presentation of this publication is a special feature of the Human Genome Project Information Web site.|