Generative AI for anxiety, depression and psychosis is sponsored by Wellcome Trust. This programme funds fundamental research on using generative AI to improve the measurement or treatment of anxiety, depression and psychosis.

Generative AI for anxiety, depression and psychosis - Grant Funding | Wellcome This website will not work correctly in Internet Explorer 11 and it is strongly recommended that you upgrade to an up-to-date browser. Internet Explorer 11 will go out of support and be retired on June 15, 2022. For more information on upgrading please see browser-update.

org . Lead applicant career stage : Early-career researcher , Mid-career researcher , Established researcher Administering organisation location : Anywhere in the world (apart from mainland China) Funding amount : Up to £3 million per award Funding duration : Up to 2 years. Participants will join an accelerator stage for the first 4 months, which is not a grant.

Then, they will have the opportunity to apply for grant funding for up to 2 years. Upcoming application stage Calculating next key date… Application process timeline The application process for this programme is in two phases: a four-month accelerator stage followed by a separate funding call. Only participants selected for the accelerator stage will be eligible to apply for the funding call.

The accelerator stage will support teams selected by Wellcome to conduct pilot studies to develop high-quality research proposals for the funding call. Wellcome will run the application and selection process for the accelerator and Neuromatch will run this accelerator stage on behalf of Wellcome.

The accelerator stage is not a grant; instead, the support will include: access to upskilling workshops funds for the development of pilot studies Find out more about what we offer in the accelerator phase. The funding call will open in August 2025, and only teams participating in the accelerator stage will be eligible to apply. To submit a proposal for funding, you must apply for the accelerator first and be selected .

We will only accept applications from teams, not individual researchers. To find a team to apply with or to recruit additional team members, you can use the matchmaking service offered by Neuromatch . This service will close on 14 April 2025.

Be of an appropriate size for the proposed research. Teams must consist of at least two applicants (including the lead applicant) and, at most, eight applicants (excluding collaborators). Have expertise in mental health research.

Teams intending to work on measurement problems should have demonstrable expertise in mental health measurement; those intending to work on improving intervention should have demonstrable expertise with developing mental health interventions. Have expertise working with generative AI.

Teams should have relevant technical expertise for the methods they intend to use in their research, and the degree of technical expertise required will scale with the technical sophistication of the methods proposed. Have clinical expertise if their proposal would likely benefit from clinical input, for example research on AI-clinician collaboration.

Include lived experience expertise of anxiety, depression or psychosis, as appropriate for the proposed research. Read our guidance on how to embed lived experience expertise in mental health research . Have expertise in AI ethics.

Expertise should span the research pipeline including model development, model refinement, evaluation and deployment. The team must also include knowledge of the unique ethical considerations relevant to the use of generative AI in mental health and experience relevant to their intended area of research. Have software engineering expertise if proposals are likely to include significant engineering complexity.

Have experience in people and research management, as appropriate for their career stage. Have the experience or the necessary support in place, needed to lead and drive a collaborative, large-scale research proposal. Have experience of, or demonstrate commitment to, effectively leading a team that embeds lived experience expertise as relevant to the research project.

Have a permanent, open-ended, or long-term rolling contract (or the guarantee of one) for the duration of the full award (accelerator and funding call). The contract should not be conditional on receiving this award. Lead applicants with less than two years remaining on their contract at the point of application must have secured their next position at an eligible organisation and provide a letter of support from them.

Be able to commit sufficient time to participate in the accelerator and deliver the proposed activities. higher education institution not-for-profit or non-governmental research organisation non-academic healthcare organisations charity or social enterprise commercial organisation (which includes sole traders or a self-employed person’s business) They can be at any career stage and come from any relevant discipline.

Be essential for the delivery of the project and make a significant contribution, for example, in designing the proposed research and leading a specific component of the project. Have a guarantee of workspace from their administering organisation for the duration of their commitment to the proposal. They do not need to have a permanent, open-ended, or long-term rolling contract.

Be able to commit sufficient time to participate in the accelerator and deliver the proposed activities. You can involve collaborators in your proposal. Collaborators support the delivery of the project but don't lead on a specific component of the research.

For example, collaborators could support by: providing technical, clinical or subject-matter expertise providing access to tools or resources, such as: being in organisations led by or working in collaboration with lived experience experts Collaborators do not have to meet eligibility requirements. They are not required to give a minimum research time commitment.

In your application for the accelerator stage, you will need to confirm that you have contacted your proposed collaborators and they are willing to participate. Collaborators do not need to confirm their participation themselves. Read about the different applicant roles at Wellcome .

If you’ve spent time away from research Career breaks, parental leave, sick leave You can apply if you have spent time away from research (for example, for a career break, parental leave or long-term sick leave). We will take this into consideration when reviewing your application. For retired researchers, host organisations must provide space and resources for the duration of the award.

Lead applicants and coapplicants can be part-time. Part-time applicants should still be able to contribute sufficient time to the participate in the accelerator and deliver the proposed activities. Their part-time work should be compatible with delivering the project successfully.

You cannot apply to the accelerator or subsequent funding call if: You are already an applicant (lead or coapplicant) on another application for the accelerator stage of the programme. You have already applied for or hold the maximum number of Wellcome awards for your career stage. Find out how many Wellcome awards you can apply for or hold at one time, depending on your career stage.

You intend to carry out activities which involve the transfer of funds into mainland China . Is your organisation right for this programme? Where your administering organisation is based The administering organisation can be based anywhere in the world, apart from mainland China .

The organisation can be a: higher education institution non-academic healthcare organisation not-for-profit or non-governmental research organisation Commercial organisations are not eligible to apply as administering organisations for this call. However, coapplicants and collaborators can be based at commercial organisations. Is your research right for this programme?

Generative AI has the potential to advance how we diagnose and treat mental health problems, and to improve measurement and evaluation within the mental health field from symptoms to risks and outcomes. But the world’s leading generative models were not built for mental health.

For example, current models aren’t as good as humans at taking into account all the different ways that mental health problems can manifest, like gestures, tone of voice and facial expressions. They also lack effective memory systems that would allow for multi-session interventions. There is a lack of high-quality benchmarking and evaluation methods and datasets relevant to mental health.

And there is no established best practice for how generative models should interact with mental health professionals and end beneficiaries. Mental health conditions in scope This funding focuses on projects that investigate symptoms of anxiety, depression and psychotic disorders.

This includes: all types of anxiety and depressive disorders (including obsessive compulsive disorder and post-traumatic stress disorder) all forms of psychotic disorders (including schizophrenia, postpartum psychosis and bipolar disorder) We recognise that the current diagnostic categories are imperfect but removing all categories or creating new ones also presents difficulties.

Whilst we do not specify any particular diagnostic or classification system, we expect applicants to use a framework and measurement approach that fits the aim of their study and to provide a clear rationale for doing so.

What your accelerator proposal must consider Our Generative AI programme has two aims: Aim 1: To create or improve models and computational approaches so that they can safely and effectively perform complex tasks to help address challenges in mental health measurement and intervention. Aim 2: To produce evidence of how generative models can and/or should collaborate with mental health professionals and end beneficiaries.

Proposals must work towards one or both of these aims. They should do so in a way that leads to a positive impact on how we measure and/or treat anxiety, depression and/or psychosis. Our aim is to lay the groundwork for large and positive impact of generative AI in mental health by funding the development of a new generation of mental-health-first generative models.

Real-world deployment of generative-AI-powered solutions beyond the scope of this programme. Have a plan for evaluating and/or validating model performance, including transparent evaluation of bias against particular groups. Select appropriate, relevant and performant language models.

Have plans for monitoring and evaluating model output to appropriately mitigate risk if building solutions where models will interact directly with mental health professionals or end beneficiaries.

Consider the ethics around their project, specifically how they can: maximise benefit and minimise harm fairly distribute benefits and harms respect and respond to the expressed needs of people with lived experience ensure transparency and accountability Use, as a minimum, one or more of our recommended common measures if collecting primary data. You may also collect data using any other measure(s).

Involve lived experience expertise of anxiety, depression and/or psychosis as relevant to the research topic. Lived experience expertise should be representative of the target population and demonstrate that the proposal aligns with the priorities of the communities the team intends to serve. Lived experience involvement Proposals must involve lived experience expertise.

We recognise that there is a range of ways that research teams can involve and collaborate with lived experience experts including as coapplicants and collaborators. We are open to any methods of involvement that teams choose, but lived experience experts must be involved in the most appropriate and ethical ways to inform multiple aspects and stages of the project.

Lived experience experts are not research participants and their input should not be limited to user testing. Lived experience experts should be engaged as colleagues who use their knowledge and expertise to inform the strategic direction, design and delivery of the research, including in leadership and governance roles.

What your research proposal can include Approaches to tackle Aim 1 could include (but are not limited to): developing new model capabilities (like memory or reasoning) or improving existing ones in a way that would lead to more competent models when applied to mental health improving model capabilities that work with sensitive data (like electronic health record data) improving model capabilities for securing sensitive data while creating new datasets for model tuning (for example securing data based on differential privacy) generating and using new datasets to specialise generative models for application in mental health (through fine-tuning, retrieval-augmented generation, few-shot learning or other functionally similar methods) developing entirely new methods that improve model performance when applied to mental health, for example, new tokenisation approaches or new prompting strategies developing new, small generative models that are specialised for mental health developing new methods and datasets to evaluate and benchmark model performance when applied to mental health measurement Approaches to tackle Aim 2 could include: comparing the strengths and weaknesses of AI models and human experts to identify how they could best complement each other developing methods to arbitrate between generative models and expert humans for specific tasks in mental health investigating how to present model output and receive input from end beneficiaries and mental health professionals, including the design of new user interfaces and modes of engagement developing new methods, approaches, and datasets to evaluate the quality and safety of human-AI interaction in mental health developing ways to identify biases and limitations (for example cultural biases or limitations) These lists are not exhaustive, and other innovative approaches to tackling these aims are welcome.

Research that is not right for this programme Whatever your approach, this programme will not fund any real-world deployment or application of generative AI. Proposals are welcome to test models, measures and interventions with consented research participants who may have been diagnosed with mental health problems, within controlled experiments (including any necessary ethical and legal approvals).

But proposals must not deploy their solutions for broader access by the general population within the lifetime of this award. This includes controlled use through healthcare systems or free use through online delivery – both are beyond the scope of awards made through this call. Working with foundation models This programme will support researchers applying generative AI models to mental health.

Access to foundation models for the purpose of the programme can take multiple routes: Pre-existing partnership with industry: Researchers with existing connections to organisations building foundation models are welcome to apply in partnership with that organisation.

New partnership with industry: If you do not have existing connections to a foundation model provider and believe your research would benefit from partnership, you may wish to collaborate with Google during the accelerator programme and subsequent funded research. Wellcome has partnered with Google for this purpose.

Teams collaborating with Google may receive: support from subject matter and/or technical experts at Google Health and Google DeepMind guidance and access to models as is appropriate for specific research projects Note that requesting a collaboration with Google does not guarantee that you will work together.

You may otherwise wish to work with other model providers (or with none at all), and you should make your decision based on what would most benefit your research project. No partnership with industry: You don’t need to partner with a model provider to be funded through in this programme.

You are welcome to use the many existing public mechanisms of accessing foundation models in your research, and to do so independently of model providers. Clarification: working with specific model providers, Google included, will not confer any advantage to your application. You should have a clear justification for why the specific foundational models you intend to use in your research are appropriate.

The goal of the accelerator stage of the programme is to help teams produce high-quality research proposals at the intersection of AI and mental health.

Teams accepted into the accelerator will gain critical support, including: comprehensive support for research proposal development workshops on designing pilot experiments and project, code, and data management for collaborative research projects funding for pilot research and/or pilot model development funding to cover contributions of team members that cannot join the accelerator as part of an existing salaried position collaboration opportunities with foundation model industry partners expert-led support for integrating lived experience in research ethics support to ensure responsible AI development You will not be able to ask for funds in your accelerator application.

If selected for the accelerator, you will then be able to let Neuromatch know what support you require. How to apply for the accelerator stage Applications are now open for the accelerator stage. Only teams selected for the accelerator phase will be eligible to apply for the funding call.

make sure you read the details on this page make sure your organisation can comply with our terms and conditions if awarded carefully read the application guidance which will guide you on how to fill in the application form download the additional questions template that you will need to submit along with your proposal you do not need to contact us before you write and submit your application As part of your application, you will also need to complete the additional questions template .

How accelerator stage applications are assessed An advisory committee will review applications to the accelerator stage of the programme and make recommendations to Wellcome. Wellcome will use these recommendations as a basis for final acceptance to the accelerator. In their review, the advisory committee will use the following criteria.

The team, skills, and experience (60%): The proposal will be delivered by a multidisciplinary, collaborative team. The team may (but is not required to) have partnered with foundation model providers or relevant platform holders.

The team includes an appropriate combination of individuals and organisations with the capacity, skills and experience to deliver impactful research at the intersection of mental health and generative AI and complies with the requirements set out in the who can apply section .

Potential for impact and feasibility of approach (40%): The proposal has the potential to advance the measurement of and/or intervention in anxiety, depression or psychosis.

The proposal clearly articulates: the potential for the proposed research to lead to novel ways to measure the presence, nature and/or degree depression, anxiety or psychosis, a subtype of this mental health problem or an outcome important to mental health the potential for the proposed research to lead to improvements in how measures are used or scored by integrating multimodal/more contextual information the potential for the proposed research to lead to more effective and/or more scalable therapeutic intervention in future downstream application whether the eventual impact of the proposed research is responsive to the expressed needs of people with lived experience of anxiety, depression and/or psychosis The proposed approach to delivering impact is feasible.

The approach: is appropriately justified, including a clear summary of any relevant evidence supporting the use of similar methods will deliver new or improved generative models that are better specialised for application in mental health will generate evidence on how best to design collaborations between generative AI, mental health professionals and end beneficiaries selects foundation models that are appropriate for the chosen research question and selects an appropriate range of methods to apply to those models proposes feasible methods of developing new generative models (if relevant) and ethical collection of data for training purposes will safely collect useful data for specialising and/or evaluating model performance includes robust and potentially novel methods for evaluating, benchmarking and/or validating model performance and for ensuring end user safety adequately attends to the range of ethical considerations arising in the design, evaluation and potential downstream deployment of the methods developed, including attention to social, political and cultural considerations appropriately manages any sensitive data and is compliant with all relevant data protection regulations mitigates key risks to achieving impact The funding call taking place after the accelerator stage will include additional assessment criteria on ‘Lived Experience and other Stakeholder Involvement’ and ‘Research Environment & Culture’.

We will communicate these assessment criteria in more detail to eligible teams closer to the time. Application process timeline You must submit your application by 17:00 BST on the deadline day. We don’t accept late applications.

Accelerator applications open Matchmaking expressions of interest close Accelerator application deadline Accelerator decisions announced Applications open to invited applicants 2 December 2025, 15:00 GMT Eligibility and application questions If you have a question about eligibility, what we offer or about completing the application form using Wellcome Funding, send a message to our funding information advisers .

Phone us on (0)20 7611 5757 Wellcome will not answer questions about scope of proposals. Applicants will need to use the information on this page and the support provided during the accelerator to develop funding proposals.