Detecting transcriptionally active regions using genomic tiling arrays

Table 1 Summary of sequence correction models

Model	Formalism	Number of parameters	Average R²	Average adjusted R²	Number of transcriptionally active regions
Uncorrected	NA	NA	NA	NA	47,463
GC	log I = β₀ + β_GC(N_C+ N_G)	2	0.0293	0.0284	52,384
Nucleotide-specific	log I = β₀ + β_AN_A+ β_CN_C+ β_GN_G	4	0.0412	0.0373	53,982
Bilinear	log I = β₀ + $\sum_{i = 1}^{36} δ_{i} β_{b (i)}$	41 = 36 + 4 + 1	0.0980	0.0604	61,731
Full Position-specific	log I = β₀ + $\sum_{i = 1}^{36} δ_{i, b (i)}$	109 = 36 × 3 + 1	0.1709	0.0703	71,400

Overview of the models used to relate probe sequence to signal intensity. The Full Position-specific model has the highest R² and also the highest adjusted R², indicating that overfitting is not a concern. The rightmost column shows the number of probed loci classified as transcriptionally active, which varies greatly with the sequence model used. NA, not applicable.

ISSN: 1474-760X