Background. Depression, substance use disorders, anxiety disorders, bipolar disorders and schizophrenia rank first in the global burden of mental diseases that lead to disability. They can occur across the lifespan, affecting those who suffer from it, their families and society in general. Only in Mexico, mental disorders occupy the eighth place in disability-adjusted life years and first place in years lived with disability with 14.92% prevalence. The urgent need for effective mental health care in combination with the progress of Artificial Intelligence (AI), has led to a significant increase in cross-disciplinary research work in this field. Particularly, the Natural Language Processing (NLP) community has significantly contributed to the computer-supported detection of mental disorders. At the same time, many researchers have started questioning the intelligible characteristics of such models (i.e., explainable and interpretable capabilities), especially when designed to support the initial diagnosis within a clinical context. AI supported diagnosis models, such as the one specified in this project, must fulfil high-level features: trust, robustness, transferability, fairness, and privacy. They are therefore a perfect ally tool for the specialists at the moment of analysing, diagnosing, and monitoring subjects having a mental disorder. Objectives. This project aims to investigate what aspects of patients’ language and behaviour can be effectively and efficiently modelled by very recent AI techniques in the diagnostic construct of depression disorders, considering relevant demographic variables such as cultural background (native language) and gender. More in particular, we aim to answer what type of knowledge, extracted from the data and those provided by the domain expert, can be exploited and infused into the learning process adding intelligibility capabilities to the generated models. The outcomes of this project will have an impact in the clinical context, where digital solutions are required to reduce the limitations of the healthcare systems of low- and mid-income countries. Methods. Recent neural language models require large amounts of data to be trained and successfully used in many NLP tasks. However, previous work has shown that having access to such amounts of clinical data is not feasible, resulting in opaque models unable to provide explainable or interpretable insights to experts. In addition, much previous work has focused on the analysis and detection of mental health problems in online social networks (OSN), raising issues of privacy, and transparency problems regarding the quality and veracity of the data. We will work with clinical data to model the language of subjects suffering from a depression disorder. To face the lack of large amounts of data we plan to: i)) use sustainable AI techniques, a key aspect that will facilitate adding structure and external knowledge from psycho-linguistics and psychiatric fields, and ii) work on data augmentation techniques considering the specific nature of the problem. Furthermore, we will exploit textual and acoustic modalities to enrich our proposed approaches’ generalisation capabilities. The impact within the clinical context will be validated with the hand of experts from a psychiatric hospital in Mexico.Expected results and impact. We expect to build a variety of computational models to assist experts during the diagnostic construct of depression disorders: identification of symptoms and risk factors, predictions about disease progression, and online psychometric tools. This will lead to insights w.r.t. the limitations and capabilities of neural-based approaches to mental health and well-being domains. In addition, we will learn how far we can leverage experts’ knowledge for building AI-supported diagnosis models. Finally, we will collect and label a large clinical dataset in Mexican Spanish, an important contribution to the research community worldwide. Overall, our research will impact the field by pushing toward AI tools that are more intelligible and implementable in practice, leading to a new level of innovation and progress needed in the digital mental health field.