AD subtypes

Alzheimer’s disease (AD) is a highly heterogeneous disease with diverse
clinical manifestations and rate of progression. AD subtypes
identification will improve our understanding of the underlying disease
mechanisms, enable us to predict disease trajectory and develop new
disease-modifying treatments.

Objective. To identify distinct AD subtypes from WGS data of AD individuals

Analysis plan. We will use 3000 WGS data derived from the ADSP Discovery
Case-Control Based Extension Study. We will use the available SNVs and
INDELS and infer structural variants (SVs) with our in-house
multi-caller pipelines. Rare variants will be retained for further
analysis. We will then split the dataset in training and tests set, and
use the identified set of genetic variants (i.e. SNVs, INDELS and SVs)
as input to a deep neural network (an autoencoder architecture) to learn
an unsupervised latent representation of the data. AD subtypes will be
identified within this reduced space and characterized using,
demographics and clinical data. We will then contrast each subtype with
the control groups to identify subtype relevant variants (i.e. putative
subtype biomarkers), which will be used as input features to a gradient
boosted tree model, to generate a subtype predictive model and subtype
specific features.

Planned collaboration. Each member of the team will devote effort in
specific areas of investigation, nevertheless, all the team members will
discuss, through regular meeting, individual progress and potential
challenges. In particular, Dr Coppola (Research Scientist, Department of
Pathology, Yale University, USA), together with Dr Dean Palejev
(Associate Professor, GATE Institute, Sofia University, Bulgaria) will
be involved in the deep learning model generation and validation, and
subtype identification; Dr Fredrik Johansson (Assistant Professor,
Department of Computer Science & Engineering, Chalmers University of
Technology. Sweden), will work on the supervised machine learning model;
Dr Alexander Schliep, Associate Professor, Department of Computer
Science & Engineering, University of Gothenburg, Sweden), will work on
the SVs inference.

Alzheimer’s disease (AD) is a highly heterogeneous disease with diverse clinical manifestations and rate of progression. The heterogeneity of AD has complicated both clinical trial design and outcomes, and thus the need for better models of AD, and/or better strategies for selection of participants into specic clinical trials is evident. The identication of more homogeneous disease subgroups (i.e. AD subtypes) will improve our understanding of the underlying disease mechanisms, enable us to predict disease trajectory and develop new disease-modifying treatments. We will use a comprehensive set of genetic variants in combination with deep learning algorithms to identify AD subtypes. Subtypes will be characterized using clinical and demographic data. Finally, variants specic to each cluster will be identied and used to train a predictive machine-learning model to classify new individuals.