Investigating the role of Copy Number Variants in Specific Language Impairment and identification of new candidate genes

Specific language impairment (SLI) is a complex neurodevelopmental disorder defined as an unexpected failure to develop normal language abilities for no obvious reason. Copy number variants (CNVs) are an important source of variation in the susceptibility to neuropsychiatric disorders. Therefore, a CNV study within SLI families was performed to investigate the role of structural variants in SLI. Among the identified CNVs, we focused on CNVs on chromosome 15q11-q13, recurrently observed in neuropsychiatric conditions, and a homozygous exonic microdeletion in ZNF277. Since this microdeletion falls within the AUTS1 locus, a region linked to autism spectrum disorders (ASD), we investigated a potential role of ZNF277 in SLI and ASD. Frequency data and expression analysis of the ZNF277 microdeletion suggested that this variant may contribute to the risk of language impairments in a complex manner, that is independent of the autism risk previously described in this region. Moreover, we identified an affected individual with a dihydropyrimidine dehydrogenase (DPD) deficiency, caused by compound heterozygosity of two deleterious variants in the gene DPYD. Since DPYD represents a good candidate gene for both SLI and ASD, we investigated its involvement in the susceptibility to these two disorders, focusing on the splicing variant rs3918290, the most common mutation in the DPD deficiency. We observed a higher frequency of rs3918290 in SLI cases (1.2%), compared to controls (~0.6%), while no difference was observed in a large ASD cohort. DPYD mutation screening in 4 SLI and 7 ASD families carrying the splicing variant identified six known missense changes and a novel variant in the promoter region. These data suggest that the combined effect of the mutations identified in affected individuals may lead to an altered DPD activity and that rare variants in DPYD might contribute to a minority of cases, in conjunction with other genetic or non-genetic factors.

Towards Unstructured Knowledge Integration in Natural Language Processing

In the last decades, Artificial Intelligence has witnessed multiple breakthroughs in deep learning. In particular, purely data-driven approaches have opened to a wide variety of successful applications due to the large availability of data. Nonetheless, the integration of prior knowledge is still required to compensate for specific issues like lack of generalization from limited data, fairness, robustness, and biases. In this thesis, we analyze the methodology of integrating knowledge into deep learning models in the field of Natural Language Processing (NLP). We start by remarking on the importance of knowledge integration. We highlight the possible shortcomings of these approaches and investigate the implications of integrating unstructured textual knowledge. We introduce Unstructured Knowledge Integration (UKI) as the process of integrating unstructured knowledge into machine learning models. We discuss UKI in the field of NLP, where knowledge is represented in a natural language format. We identify UKI as a complex process comprised of multiple sub-processes, different knowledge types, and knowledge integration properties to guarantee. We remark on the challenges of integrating unstructured textual knowledge and bridge connections with well-known research areas in NLP. We provide a unified vision of structured knowledge extraction (KE) and UKI by identifying KE as a sub-process of UKI. We investigate some challenging scenarios where structured knowledge is not a feasible prior assumption and formulate each task from the point of view of UKI. We adopt simple yet effective neural architectures and discuss the challenges of such an approach. Finally, we identify KE as a form of symbolic representation. From this perspective, we remark on the need of defining sophisticated UKI processes to verify the validity of knowledge integration. To this end, we foresee frameworks capable of combining symbolic and sub-symbolic representations for learning as a solution.