Research Strategy
As a leader for the CPDSE, my main focus is on the development of the center as a whole and its mission. This should remain the leader’s main priority for the duration of the center. However as the center becomes robust, evolves and co-leadership truly kicks in (see leadership principles), more effort can be shifted into my professorial research activities.
I am considering two avenues of research:
- Enabling Oligonucleotide drug discovery through data and modelling (this document below)
- Drug discovery process modelling and strategy (to be further elaborated)
Enabling Open Oligonucleotide Drug Discovery Through Data and Modelling.
The research strategy in the first years will focus on data curation, machine learning, community building, simulations. Several of the envisioned projects will have outcomes both in terms of uncovering basic science and in terms of producing tools, frameworks and models that can be used for teaching and strengthening the oligonucleotide community.
The draft research strategy will be further evolved together with current collaborators, future collaborators inside and outside of the department, and evolving opportunities for grants.
Research Strategy Timeline
Horizon 1 (1-2 yrs): Establishment phase.
Build on existing strengths, high probability of success.
OpenOligoLake
Data Curation & Organization
Curate and organize foundational oligo discovery data, systems and code to enable oligonucleotide based drugs.
- Requirement: Own students, postdocs for coding and data mining and curation
- Existing network: Consolidate already public data and code
- Building on: Center resources on data and code orchestration, increment on top current work
OpenOligoFeat
Machine Learning for Structure-Function
Featurization and ML for learning and mapping between structure and function for oligonucleotides. Apply to predict safety, activity and stability on public data sets.
- Requirement: Own students, postdocs for coding and data curation; collaboration on advanced machine learning
- Building on: Experience with hepatotox, acute neurotox and activity predictors, public datasets (from OpenOligoLake)
- Expansion: Base can be quickly and independently established, then expand through industry and tech collaborations (e.g. Roche, Novo, Silence, Abzu.ai)
OpenOligoCommunity
Pre-Competitive Partnerships
Activate and expand existing networks to build pre-competitive partnerships to enable and accelerate oligonucleotide drugs. Pursue national and EU funding.
- Challenge: Current Roche anchoring makes this more difficult
- Solution: Not-for-profit anchoring provides basis for catalyzing open sharing and diverse funding
- Goal: Facilitate public or joint ownership models or open source in precompetitive spaces
- Requirement: Outreach and ecosystem building, leading to joint grant applications
OpenNABMPKPD
Mechanistic Systems Pharmacology Models
Mechanistic systems pharmacology models for nucleic acid based medicines.
Expanding from ASO-model with PK and into other NABMs. Compartment-models for tissues, crossing barriers, entering cells, and molecular mode of action. Contrary to standard PKPD models used in drug development, this system of models would be more detailed and mechanistic including cellular uptake, stability of both pharmacophore and targeting system, molecular mechanisms of action (enzymes for oligos, transcription, translation etc for AAVs).
Benefits: - Better understanding of what to optimize - Framework to become quantitative and more precise - Simulations used for teaching - Framework to simulate and plan wetlab experiments - Guidance for dosing schemes in animal and human studies
- Requirement: Own students, collaboration on math for dynamic models
Horizon 2 (2-5yrs): Expansion phase.
Requires significant new know-how and capabilities either in the group or through tight collaborations. Additional funding necessary and expected.
Design-Build-Test-Learn Loops
To accelerate learning about oligonucleotide drugs it is essential to have fast learning loops between molecule design, synthesis, testing and learning. In this phase this should be established either through own lab footprint or through collaborations and vendors/CROs.
OpenTranscriptomeFoundationModel
Foundation Models for the Transcriptome
Given an oligonucleotide chemical structure, what are the predicted transcriptome changes? And given a desired transcriptome change (e.g. sick to healthy) what oligonucleotides will induce change? This is almost the holy grail of RNATherapy data science.
- Requirement: Development and connection to expertise on RNA-foundation models; connection to oligo perturbed transcriptome data sets (mined through OpenOligoLake or produced through collaboration)
- Expected evolution: RNASeq-methods (e.g DrugSeq or BRB-Seq) in early screening campaigns for oligonucleotides drugs
OpenDiscoveryCascadeModel
Optimization of Drug Discovery Cascades
One can define the drug discovery business process of lead identification and optimization as a series of operations that attempts to predict the in-human properties (e.g. efficacy, PK and safety) of sets of molecules. Each operation carries a cost in terms of time and money; an operation’s predictive validity is its ability to predict the outcome of a later, often more expensive, operation.
The Challenge: Discovery cascades are almost always designed by qualitative tradition, opinions and assumptions. Real discovery cascades are slow (2+ yrs) and involve many different specialists, meaning that holistic systems learning is very slow and difficult to implement.
The Vision: Create a stochastic graphical model abstraction of discovery cascades to enable: 1. More rational cascade design 2. Teaching tool—turn into a learning game with ‘engine building mechanics’ usable in university classes and pharma project/strategy teams
- Requirement: 2 students/post-docs; connection to graphical models expertise; game and workshop design expertise; industry connections for examples and parameter estimation
- Builds on: Experience with 30+ cascades; recreational game design