Grant: Discovery Grants Program - Individual
The University of British Columbia
$87,000.00
Bioinformatics for Prediction of Transcriptional Regulation Mechanisms
FY 2024-25
Summary
The capacity for cells to take on new characteristics in response to changes in their biological context is fundamental to multicellular life. From the same DNA instructions, distinct sets of genes can be expressed to allow cells to protect themselves against chemicals, interact with other cells or form structures (e.g. organs or tissues). Such changes are enabled by changing patterns of transcription, the redirection of information flow from DNA to RNA.
From a computational perspective, the study of DNA and transcription presents an ideal area for machine learning methods to be developed and applied. New techniques are constantly expanding the scale of data production related to gene regulation; increasingly delivered at single cell resolution. By determining patterns within the data, it becomes possible to associate transcriptional responses to specific conditions, to delineate segments of DNA that control gene activity levels and to begin to unravel the complexity of interactions between transcription factors (TFs) required for targeted activity.
This propsoal takes a novel machine learning approach for obtaining insights into gene regulation. Deep learning is now widely used to make predictions based on patterns in large data. We recently introduced ExplaiNN, a method based on neural additive models. ExplaiNN allows detection of sequence motifs (TF binding sites) and their relative contribution to expression (both globally across genes and locally within a gene). As deep learning progresses, researchers will increasingly wish to empower the computer to consider more attributes than DNA molecular sequence in the model construction process. If one compares human intuition to the current methods for inference of mechanisms about transcriptional regulation, one will note that we draw much of our capacity from information reported in the scientific literature and by analogy to similar systems. Existing machine learning methods draw most capacity from molecular measurements. We will introduce annotation-based characteristics of TFs into the process by building knowledge graphs for each TF based on existing annotation resources and automated extraction of characteristics from literature.
The proposal encompasses three aims:
Aim 1 - Implementation of a TF Knowledge Graph
Aim 2 - Generation of Embeddings
Aim 3 - Interpretable deep learning with TF embeddings
Utilizing molecular data and functional annotation represented in knowledge graphs, a new predictive deep learning model will be implemented. At the center of the work is the representation of TF-related knowledge in a Knowledge Graph, and the compilation of contextual information about TFs in embeddings - numerical vectors capturing the properties observed in large data, such as text or the spatial properties of the graph itself. A focus will be maintained on interpretability, allowing the relative contributions of information types to be assessed.
Program
Discovery Grants Program - Individual
Recipient
Location
British Columbia, Canada
Recipient Type
Academia
Type
Grant
Source
NSERC's Awards Data
As published
Application id
RGPIN-2024-06783
Name nom
Wasserman, Wyeth
Department département
Medical Genetics
Organization id
2
Institution établissement
The University of British Columbia
Province en
British Columbia
Province fr
Colombie-Britannique
Country en
CANADA
Country fr
CANADA
Fiscal year exercice financier
2024
Competition year année de concours
2024
Award amount
87000
Program id
RGPIN
Program name en
Discovery Grants Program - Individual
Program name fr
Programme de subventions à la découverte - individuelles
Group en
DISCOVERY RESEARCH
Group fr
RECHERCHE AXÉE SUR LA DÉCOUVERTE
Committee code
1501
Committee name en
Genes, Cells and Molecules
Committee name fr
Gènes, cellules et molécules
Area of application code
1201
Area of application group en
Advancement of knowledge
Area of application group fr
Avancement des connaissances
Area of application en
Life sciences (including biotechnology)
Area of application fr
Sciences de la vie (y compris biotechnologie)
Research subject code
5303
Research subject group en
Molecular biology
Research subject group fr
Biologie moléculaire
Research subject en
Gene regulation
Research subject fr
Régulation des gènes
Field of research list names en
Not available
Field of research list names fr
Non disponible
Socioeconomic objective list names en
Not available
Socioeconomic objective list names fr
Non disponible
Application title
Bioinformatics for Prediction of Transcriptional Regulation Mechanisms
Keywords
transcription; genome; machine learning; dna; transcription factor; pattern discovery; deep learning; cell; differentiation; chromatin
Application summary
The capacity for cells to take on new characteristics in response to changes in their biological context is fundamental to multicellular life. From the same DNA instructions, distinct sets of genes can be expressed to allow cells to protect themselves against chemicals, interact with other cells or form structures (e.g. organs or tissues). Such changes are enabled by changing patterns of transcription, the redirection of information flow from DNA to RNA.
From a computational perspective, the study of DNA and transcription presents an ideal area for machine learning methods to be developed and applied. New techniques are constantly expanding the scale of data production related to gene regulation; increasingly delivered at single cell resolution. By determining patterns within the data, it becomes possible to associate transcriptional responses to specific conditions, to delineate segments of DNA that control gene activity levels and to begin to unravel the complexity of interactions between transcription factors (TFs) required for targeted activity.
This propsoal takes a novel machine learning approach for obtaining insights into gene regulation. Deep learning is now widely used to make predictions based on patterns in large data. We recently introduced ExplaiNN, a method based on neural additive models. ExplaiNN allows detection of sequence motifs (TF binding sites) and their relative contribution to expression (both globally across genes and locally within a gene). As deep learning progresses, researchers will increasingly wish to empower the computer to consider more attributes than DNA molecular sequence in the model construction process. If one compares human intuition to the current methods for inference of mechanisms about transcriptional regulation, one will note that we draw much of our capacity from information reported in the scientific literature and by analogy to similar systems. Existing machine learning methods draw most capacity from molecular measurements. We will introduce annotation-based characteristics of TFs into the process by building knowledge graphs for each TF based on existing annotation resources and automated extraction of characteristics from literature.
The proposal encompasses three aims:
Aim 1 - Implementation of a TF Knowledge Graph
Aim 2 - Generation of Embeddings
Aim 3 - Interpretable deep learning with TF embeddings
Utilizing molecular data and functional annotation represented in knowledge graphs, a new predictive deep learning model will be implemented. At the center of the work is the representation of TF-related knowledge in a Knowledge Graph, and the compilation of contextual information about TFs in embeddings - numerical vectors capturing the properties observed in large data, such as text or the spatial properties of the graph itself. A focus will be maintained on interpretability, allowing the relative contributions of information types to be assessed.