Identification of transcription start sites via distribution of A/T-singletons

Transcription start sites (TSSs) are crucial information that determines exact location of genes. However, identifying TSSs in vitro is costly and time consuming. Therefore, there are many attempts to predict TSSs in silico, but they were low in accuracy. Herein, we observed that the distribution of A/T-singletons in the whole genome can be employed to detect the presence of TSS. We found that the distribution pattern is clearly distinct between regions with TSS and without TSS. To assess whether the distribution of A/T-singletons can help detect TSS, we developed a two-step algorithm to detect the specific distribution pattern. Our method was evaluated in terms of sensitivity, specificity and accuracy. The results show that the distribution of A/T-singletons is a useful feature for identifying TSSs. However, using this feature alone is not sufficient to identify all TSSs correctly. Combining this feature with other existing methods should improve their efficacy significantly.