DNA methylation analysis by sequencing is becoming increasingly popular, yielding methylomes at single-base pair and single-molecule resolution. It has tremendous potential for cell-type heterogeneity analysis using intrinsic read-level information. Although diverse deconvolution methods were developed to infer cell-type composition based on bulk sequencing-based methylomes, systematic evaluation has not been performed yet. Here, we thoroughly benchmark six previously published methods: Bayesian epiallele detection, DXM, PRISM, csmFinder+coMethy, ClubCpG and MethylPurify, together with two array-based methods, MeDeCom and Houseman, as a comparison group. Sequencing-based deconvolution methods consist of two main steps, informative region selection and cell-type composition estimation, thus each was individually assessed. With this elaborate evaluation, we aimed to establish which method achieves the highest performance in different scenarios of synthetic bulk samples. We found that cell-type deconvolution performance is influenced by different factors depending on the number of cell types within the mixture. Finally, we propose a best-practice deconvolution strategy for sequencing data and point out limitations that need to be handled. Array-based methods—both reference-based and reference-free—generally outperformed sequencing-based methods, despite the absence of read-level information. This implies that the current sequencing-based methods still struggle with correctly identifying cell-type-specific signals and eliminating confounding methylation patterns, which needs to be handled in future studies.

White matter hyperintensities (WMH) appear as regions of abnormally high signal intensity on T2-weighted magnetic resonance image (MRI) sequences. In particular, WMH have been noteworthy in age-related neuroscience for being a crucial biomarker for all types of dementia and brain aging processes. The automatic WMH segmentation is challenging because of their variable intensity range, size and shape. U-Net tackles this problem through the dense prediction and has shown competitive performances not only on WMH segmentation/detection but also on varied image segmentation tasks. However, its network architecture is high complex. In this study, we propose the use of Saliency U-Net and Irregularity map (IAM) to decrease the U-Net architectural complexity without performance loss. We trained Saliency U-Net using both: a T2-FLAIR MRI sequence and its correspondent IAM. Since IAM guides locating image intensity irregularities, in which WMH are possibly included, in the MRI slice, Saliency U-Net performs better than the original U-Net trained only using T2-FLAIR. The best performance was achieved with fewer parameters and shorter training time. Moreover, the application of dilated convolution enhanced Saliency U-Net by recognizing the shape of large WMH more accurately through multi-context learning. This network named Dilated Saliency U-Net improved Dice coefficient score to 0.5588 which was the best score among our experimental models, and recorded a relatively good sensitivity of 0.4747 with the shortest training time and the least number of parameters. In conclusion, based on our experimental results, incorporating IAM through Dilated Saliency U-Net resulted an appropriate approach for WMH segmentation.

2White matter hyperintensities (WMH) appear as regions of abnormally high signal intensity 3 on T2-weighted magnetic resonance image (MRI) sequences. In particular, WMH have been 4 noteworthy in age-related neuroscience for being a crucial biomarker for Alzheimer's disease 5 and brain aging processes. However, the automatic WMH segmentation is challenging because 6 of the variable intensity range, size and shape. U-Net tackled this problem through the dense 7 prediction and showed competitive performances on not only WMH segmentation/detection but 8 also on varied image segmentation tasks, but it still accompanies a high complexity of the network 9 architecture. In this study, we propose to use Saliency U-Net architecture and irregularity age 10 map(IAM) to decrease the U-Net complexity without a performance loss. We trained Saliency 11 U-Net using both T2-FLAIR MRI sequence and IAM. Since IAM guides where irregularities, in 12 which WMH is possibly included, exist on the MRI slice, Saliency U-Net performs better than the 13 original U-Net trained only using T2-FLAIR. The better performance was achieved with fewer 14 parameters and shorter training time. Moreover, the application of dilated convolution enhanced 15 Saliency U-Net to recognise the shape of large WMH more accurately by learning multi-context 16 on MRI slices. This network named Dilated Saliency U-Net improved Dice coefficient score to 17 0.5588 which is the best score among our experimental models, and recorded a relatively good 18 sensitivity of 0.4747 with the shortest train time and the least number of parameters. In conclusion, 19 based on the experimental results, incorporating IAM through Dilated Saliency U-Net resulted an 20 appropriate approach for WMH segmentation. 21 22 25progression (Gootjes et al., 2004). Higher volume of WMH has been found in brains of AD patients 26 compared to age-matched controls, and the degree of WMH has been reported more severe for senile onset 27 AD patients than presenile onset AD patients (Scheltens et al., 1992). Furthermore, WMH volume generally 28 increases with the advance of age (Raz et al., 2012; Jagust et al., 2008). Due to their clinical importance, 29 various machine learning approaches have been implemented for the automatic WMH segmentation 30 (Admiraal-Behloul et al., 2005; Bowles et al., 2017).31 LOTS-IM algorithm is an unsupervised algorithm for detecting tissue irregularities, that successfully has 32 been applied for segmenting WMH regions on brain T2-FLAIR images (Rachmadi et al., 2019). Without 33 any ground-truth segmentation, this algorithm produces a map which describes how much each voxel is 34 irregular compared with an overall area. This map is usually called 'irregularity map' (IM) or 'irregularity 35 age map (IAM)'. The concept of this map was firstly suggested in the field of computer graphics to calculate 36 1 Yunhee Jeong et al. DSUNet WMH segmentation using IAM pixel-wise age values indicating how weathered/damaged each pixel is compared to the overall texture 37 pattern of an ...

DNA methylation sequencing is becoming increasingly popular, yielding genome-wide methylome data at single-base pair resolution through the novel cost- and labor-optimized protocols. It has tremendous potential for cell-type heterogeneity analysis, particularly in tumors, due to intrinsic read-level information. Although diverse deconvolution methods were developed to infer cell-type composition based on bulk sequencing-based methylomes, their systematic evaluation has not been performed so far. Here, we thoroughly review and evaluate five previously published deconvolution methods: Bayesian epiallele detection (BED), PRISM, csmFinder + coMethy, ClubCpG and MethylPurify, together with two array-based methods, MeDeCom and Houseman as a comparison group. Sequencing-based deconvolution methods consist of two main steps, informative region selection and cell-type composition estimation. Accordingly, we individually assessed the performance of each step and demonstrated the impact of the former step upon the performance of the following one. In conclusion, we demonstrate the best method showing the highest accuracy in different samples, and infer factors affecting cell-type deconvolution performance according to the number of cell types in the mixture. We found that cell-type deconvolution performance is influenced by different factors according to the number of components in the mixture. Whereas selecting similar genomic regions to DMRs generally contributed to increasing the performance in bi-component mixtures, the uniformity of cell-type distribution showed a high correlation with the performance in five cell-type bulk analyses.