Methylation, as an important epigenetic regulation of methylation genes, plays an important role in the regulation of some biological processes. Although the traditional methods of detecting methylation in biological experiments are constantly improving, with the development of artificial intelligence, using deep learning and machine learning methods to detect methylation has gradually become a new trend. Traditional machine learning based methods rely too much on manual feature extraction. Although there are some deep learning methods to study methylation, most network architectures extract fewer features due to the simple network structure. We propose a bottleneck network based on attention mechanism, and use some new methods to ensure that the deep network can learn more effective features and minimize overfitting to get more significant prediction results. The model uses three coding methods to encode the original DNA sequence, and then uses the feature fusion based on the attention mechanism to get the best fusion method. The results show that MLACNN is superior to the previous methods and can obtain more satisfactory performance.