Accelerated development of DIPG selective nucleic acid formulations using machine learning methods

Diffuse Intrinsic Pontine Glioma (DIPG) is an aggressive and malignant rare paediatric brain tumour found in the brainstem which is currently uncurable. The diffuse nature of the malignant glial cells and challenges of administering therapeutics to the pons mean that drug formulations should ideally be optimally designed to be uptaken and processed by DIPG cells compared to the surrounding local tissue. One promising therapeutic modality for a range of cancers are siRNAs, which can transiently knockdown overexpressed genes to suppress oncogenes (e.g. related to drug resistance, proliferation, mutation etc) hence may be vital therapeutic options in DIPG treatment. The most common approach to stabilise and deliver nucleic acid therapeutics are lipid nanoparticles (LNPs), as used in Onpattro and the two licensed COVID-19 mRNA vaccines. Optimising such complex delivery systems with a vast formulation space to produce the desired nanoparticle properties (size, charge, surface chemistry) and to be DIPG specific would be highly resource intensive with a high-throughput screening only approach.
This project aims to develop and use machine learning methodologies to optimise the composition and physicochemical properties of LNPs to improve selectivity and siRNA delivery efficiency towards DIPG cells. We will use established literature datasets and high throughput formulation screens to develop and validate a predictive model between the LNP composition and expected physical properties of the LNPs. The model will then be further expanded to predict uptake and transfectability of siRNA with a specified LNP composition in established reporter lines and novel DIPG reporter cells. The final aim of this work will then be to combine these models with established DIPG transcriptomic data sets to produce models which can examine if genomic/transcriptomic profiles can be used as markers for compatibility with specific LNP formulations.
The ambition of this work will be to develop an in silico tool that enables dialling in a set of LNP input parameters that will be able to predict physicochemical outputs such as particle size, likely transfectability/selectivity towards DIPG cells then to merge these with genomic and transcriptomic profiles to predict delivery efficiency in any cell/tissue types using a specified LNP formulation.