DNA sequencing using the fluoresence based Sanger method comprises
interpretation of a sequence of signal peaks of varying size whose colour
indicates the presence of a base. We have established that the ability
to predict the variations effectively makes available novel error correction
information which will improve sequencing efficacy. Our experiments so
far have used basic models of the Sanger reaction chemistry and machine
learning techniques. These have enabled us to make base calls just using
context information, specfically ignoring the peak data at the base calling
position. The 80% success rate of our blind experiments is striking,
and will be improved by a more accurate model of trace behaviour. To
this end, and to integrate the information into mainstream basecalling,
we wish to develop an enzyme kinetics model susceptible to calibration of
its component rates such that trace data can be accurately predicted. We
describe DNA sequencing trace data, outline the trace prediction problem
requirements on the model, and discuss model construction and calibration
issues.