Deep Learning for Inference from Biological Data

随着我们在基因组、表观遗传、基因表达和种群水平上进行各种生物学测量的能力迅速提高，生物界普遍认识到，数学和计算模型是能够理解分子相互作用和遗传编码信息如何转化为生物功能以及如何将这种理解转化为帮助治疗复杂疾病的必要工具。尤其令人感兴趣的是基因在转录组和蛋白质组不同水平上的表达的测量，这种映射是如何调节的，以及DNA序列中编码的哪种信息是它的决定因素。我们一直在ECS和医学院之间在该领域建立了强有力的合作，迄今已有：(I)用于检测翻译后调控蛋白质的异常值检测算法(发表在《生物信息学》杂志上)，该算法后来成为KTP成功竞标的主题；(Ii)在上述基础上发表的一篇论文(发表在《免疫学杂志》上)，该论文研究了映射到调节蛋白质水平的微小RNA的新数据；(Iii)关于翻译调控主题的手稿(提交给核酸研究迫在眉睫)以及(Iv)本轮向BBSRC提交的赠款(目前正在审查中)。这个项目将在这个丰富的领域，重点放在两个我们迄今尚未解决的领域：(I)剪接的作用及其调节；(Ii)从序列水平提取信息，重点放在基因组非编码区的调控元件上。我们预计测量的特征和序列派生的特征的组合将有助于做出准确的推断，并将开发从深度学习技术中学习的高级表示法来实现这一点。

With the rapid increase in our ability to make various measurements about biology at the genomic, epi-genetic, gene expression and population levels, there is widespread acknowledgement within the biological community that mathematical and computational modelling is a necessary tool to be able to understand how molecular interaction and genetically encoded information translates into biological function, and how to translate such understanding to help in the treatment of complex diseases. Of particular interest is the measurement of gene expression at the different levels of transcriptome and proteome, how this mapping is regulated and what kind of information encoded in the DNA sequence are determinants of it. We have been building a strong collaboration between ECS and the School of Medicine in this domain and so far have: (i) an outlier detection algorithm (published in Bioinformatics) for detection of post-translationally regulated proteins, which later formed the subject of a successful KTP bid; (ii) a publication that builds on the above (in the Journal of Immunology) looking at new data that is mapped to micro RNAs regulating protein levels; (iii) A manuscript on the subject of translation regulation (submission to Nucleic Acids Research imminent) and (iv) a grant submission this round to BBSRC (currently under review). This project will be in this rich domain, focusing on two areas we have hitherto not addressed: (i) the role of splicing and its regulation; and (ii) extracting information from the sequence level with focus on regulatory elements in the non-coding regions of the genome. We expect a combination of measured and sequence-derived features to be beneficial in making accurate inferences and will be developing advanced representations learned from deep learning techniques to achieve this.