*2.3. Sequence CNNs.*

Figure 1 shows different CNN models for sequence data after splitting the data by chromosomes to prevent non-independent training, validation, and test data. We conducted an extensive evaluation of CNNs with varying structures and parameters (Table 1 and Table S2). We first tested the performance of the CNN in Zhuang, et al. [14], and then added an attention module to utilize the sequence structure and focus only on the middle part of the input sequence [26]. Moreover, we tested on the use of the Residual Neural Network (ResNet) architecture for the data (Table 2) [27,28].


**Table 1.** Parameters for convolutional neural network (CNN) models.
