1 of 16

AI in Science: ��Topological Data Analysis for AI-based Personalised Dementia Treatments

Stockholm AI�The Park Forskaren, 27/01/2026

​

​

Belén García Pascual

2 of 16

About me

​

  • Bachelor in mathematics from Complutense University of Madrid, Spain

​

  • Master in mathematics (topology) from University of Bergen, Norway

​

  • PhD in biomathematics from University of Bergen, Norway

​

  • Industry internship in generative AI in healthcare at DNV, Oslo, Norway

​

  • Postdoc at KTH (Mathematics) and Karolinska Institutet (Neurobiology)

Funded by Digital Futures

3 of 16

Topological Data Analysis for AI-based Personalised Dementia Treatments

4 of 16

Goal

To personalise drug prescriptions in patients with dementia by finding subgroups of patients with common clinical and biological profiles.

​

Medication repurposing

​

Data: Electronic Healthcare Records

  1. The Swedish Dementia Registry (SveDem)
    • Over 100 000 patients
  2. American National Alzheimer’s Coordinating Center

(NACC)

    • Over 54 000 patients, 2000 features/columns

​

​

Topological Data Analysis for AI-based Personalised Dementia Treatments

5 of 16

Topological Data Analysis for AI-based Personalised Dementia Treatments

Working with two cohorts:

​

  • There are big differences between each dataset
  • However, main patterns in the dementia disease appear in both

​

  • Models and analyses are developed using one cohort and independently validated using the other more robust findings

​

6 of 16

Topological Data Analysis for AI-based Personalised Dementia Treatments

7 of 16

Topological Data Analysis for AI-based Personalised Dementia Treatments

Topology

​

  • Abstract geometry
  • Studies the shapes of a space

​

  • A mug and a donut are topologically equal because they have the same shape (same number of holes)

​

​

​

​

​

Astronomy Nuclear Physics Info https://astronuclphysics.info/Gravitace3-1.htm

​

  • A circumference and a disk are not topologically equal (one hole vs no holes)

​

8 of 16

Topological Data Analysis for AI-based Personalised Dementia Treatments

Topological data analysis

​

  • Topology applied to data
  • We imagine that the data is a finite sample of points from an underlying space

​

​

​

​

​

​

​

​

​

​

​

​

  • By studying shapes in the space, we can learn about the structure and characteristics of the data
  • Useful in high-dimension, when the data have many features

​

Feature A

Feature B

Data points (rows in the dataset)

Feature A

Feature B

1

2

1.2

2.2

4

4.5

(1,2)

(1.2, 2.2)

(4.5, 4)

2.5

0.1

(2.5, 0.1)

Dataset

Underlying space

Structure of the data

No observations

in this region

Denser region

More sparcity =

data is little related,

distinguisable groups

9 of 16

AI in Science: topological methods across the sciences

​

​

  • In psychosis:

M. Fournier, et. al., Topology predicts long-term functional outcome in early psychosis. Molecular Psychiatry. 2021

​

  • In cancer biology:

Stolz, B. J.,et al. Relational Persistent Homology for Multispecies Data with Application to the Tumor Microenvironment. Bulletin of Mathematical Biology. 2024.

Aukerman A, et al. Persistent homology based characterization of the breast cancer immune microenvironment: a feasibility study. International Symposium on Computational Geometry. 2020.

​

  • In neuroscience:

Nielson JL, et al. Topological data analysis for discovery in preclinical spinal cord injury and traumatic brain injury. Nat Commun. 2015.

Yonggang S, et al. Persistent Reeb Graph Matching for Fast Brain Search. Machine Learning in Medical Imaging. 2014.

​

  • In genomics:

Masoomy H, et al. Topological analysis of interaction patterns in cancer-specific gene regulatory network: persistent homology approach. Sci Rep. 2021.

Benjamin K, et al. Homology of homologous knotted proteins. J R Soc Interface. 2023.

10 of 16

Topological Data Analysis for AI-based Personalised Dementia Treatments

​

11 of 16

Topological Data Analysis for AI-based Personalised Dementia Treatments

​

​

  • Topology benefits from AI-based standard methods

​

    • Dimensionality reduction techniques
    • Clustering algorithms

12 of 16

Topological Data Analysis for AI-based Personalised Dementia Treatments

  • High dimensional and complex data dimensionality reduction: visualise in the form of a graph/network

​

  • We want to find subgroups of patients with shared profiles cluster patients

​

  • The nodes/vertices in the graph are groups of patients that are similar
  • The edges in the graph are relations between patients

​

13 of 16

Topological Data Analysis for AI-based Personalised Dementia Treatments

Biomarkers for Alzheimer’s disease:

  • Amyloid-beta 1-42 low for advanced disease
  • P-tau181 high for advanced disease
  • T-tau high for advanced disease

14 of 16

Topological Data Analysis for AI-based Personalised Dementia Treatments

Numerical proxy for cognition:

• Mini-Mental State Examination (MMSE) score

​

15 of 16

Topological Data Analysis for AI-based Personalised Dementia Treatments

Still a work in progress!

​

​

Aims:

​

  • Robust model of subgroups of patients

​

  • Predictive of disease outcomes

​

  • Tailoring of specific medications to specific patients

​

  • Applying different approaches to our data (like considering time series) to validate results

16 of 16

Thanks!

Join work with:

​

​

​

​

​

​

​

​

​

​

​

​

Email: belengp@kth.se

Home page: belen.no

LinkedIn: Belén García Pascual

​

​

​

​

​

​

​

​

​

​

Martina Scolamiero

Assistant Professor

KTH

Saikat Chatterjee

Professor

KTH

Sara García Ptacek

Assistant Professor

KI