Foundational Models for Personalised Mental Health
Issued Date
2026-01-01
Resource Type
ISSN
18674534
eISSN
18674542
Scopus ID
2-s2.0-105039513544
Journal Title
Adaptation Learning and Optimization
Volume
28
Start Page
73
End Page
85
Rights Holder(s)
SCOPUS
Bibliographic Citation
Adaptation Learning and Optimization Vol.28 (2026) , 73-85
Suggested Citation
Tan G.C.Y., Wang B.Z., Tan H.M., Pe L.S., Yuen A.K.P., Mu Y., Lee B., Huixian S.L., Tan E.S., Fong K.V., Keppo J., Goh H.L., Dai P. Foundational Models for Personalised Mental Health. Adaptation Learning and Optimization Vol.28 (2026) , 73-85. 85. doi:10.1007/978-3-032-12362-6_6 Retrieved from: https://repository.li.mahidol.ac.th/handle/123456789/116954
Title
Foundational Models for Personalised Mental Health
Corresponding Author(s)
Other Contributor(s)
Abstract
Mental health disorders are heterogenous in presentation and treatment response. For example, only one third of patients started on an antidepressant will achieve remission and each trial of medication can take several weeks. Additionally side effects and the development of chronic conditions such as diabetes or high cholesterol are common. We discuss the potential application of foundation models as developed from electronic medical records (EMRs), large language models (LLMs) and for pharmacogenetics drawing potential links and applications in mental health. In terms of EMRs, the concept of a patient representation has been used across applications such as disease prediction and personalised treatment. These approaches have been applied in mental health to label diseases such as depression and bipolar disorder as well as to predict suicide in risk assessment. We discuss a range of applications for LLMs, from supporting the preprocessing of EMRs for FEMRs, therapy support through transcription and assessment and patient monitoring, and psychoeducation. We discuss the potential applications of biomedical foundation models to precision medicine with pharmacogenetics. Finally, we touch on ways of integrating broad sources of data and outputs from various models.