Skip to content

Impact case study

Commercialisation of AI-based Algorithms for Faster In Silico Drug and Diagnostic Biomarker Discovery

Unit(s) of assessment: Allied Health Professions, Dentistry, Nursing and Pharmacy

Research theme: Health and Wellbeing

School: School of Science and Technology

Impact

New approaches for studying disease systems at the genomic, epigenetic, proteomic and metabolomic levels are continually developed to predict responses to therapy or identify new drug targets. The challenges in analysing large molecular datasets lie in their volume, resolution and complexity, and the quality of the data, which can lead to false discovery and overfitting (i.e. falsely elevated diagnostic performance).

Research at NTU, led by Professor Graham Ball over 18 years, has developed a novel Machine Learning systems approach to in silico biomarker discovery. Based on Artificial Neural Networks (ANNs), Ball’s computational algorithms facilitate analysis of large complex omics datasets, to identify non-linear biomarkers associated with clinical features concordant across multiple datasets. The algorithms study interactions between nodes within a single disease pathway and identify the most influential sets of biomarkers in a given biological system.

Ball has developed patented Artificial Neural Network based techniques to facilitate extraction of aetiological meaning from complex omics data repositories. NTU spin-out Intelligent OMICS has licenced and commercialised Ball’s patented techniques, generating £230k revenues and attracting £2m finance. Intelligent OMICS has helped secure US Defence Threat Reduction Agency contracts with DSTL worth £4.5m to enable rapid diagnosis of infection, particularly trauma induced sepsis, impacting on healthcare in the military. Intelligent OMICS incorporation of AI methods into drug discovery process enabled Cumulus Oncology’s successful £1.7m fund raise. Low-cost diagnostic panels have been shown to manage COPD by Mologic, benefitting clinical trial patients, and 56 latent TB cases identified during trials with Wuhan Pulmonary Hospital (China) enabled these patients to receive early treatment.

Research background

The initial concept was to utilise ANNs with a constrained architecture enabling them to discover robust biomarkers from complex omics data. Ball’s group was the one of the first to mine mass spectrometry data using an ANN approach; the first study identified biomarker ions that accurately differentiated between two common types of brain tumours, astrocytoma and glioblastoma. These ANN algorithms were further developed to model public molecular data repositories like the Gene Expression Omnibus and The Cancer Genome Atlas in order to identify robust molecular diagnostic panels from the whole transcriptome, concordant across multiple data sets.

This approach provided a rank order of the molecular biomarkers associated with a given clinical question, overcame the issues of false discovery and overfitting, and improved the predictive performance for unseen cases from disease populations. In a collaborative study with Nottingham University Hospitals Trust (NUH), it was used to validate a gene expression profile for the detection of clinical outcomes in breast cancer patients, having reduced a 70-gene signature to just nine genes capable of accurately predicting distant metastases. Patents for the ANN data methods were filed in 2009, and a spinout company Intelligent OMICS was launched to exploit the algorithmic IP and the methods.

Further work extended the rank order concept to undertake a meta-analysis over multiple questions and multiple data sets. Ball developed a new algorithm that utilised ANN-based network inference (ANNi) methods to model networks of molecular interactions between context-specific biomarkers, for a given pathway or disease, based on whole transcriptomic data. By identifying associations between molecules in a pathway and the remaining molecules in the whole transcriptome, new pathway features for a given disease could be identified and analysed to find the most influential molecular drivers in a given system.

These new approaches were applied in a breakthrough study, with NUH, Universities of Nottingham, Cambridge and Auckland and China’s Northeast Normal University, analysing the transcriptomic gene expression levels associated with multiple proliferation features in over 4500 breast cancer cases. Published in The Lancet Oncology, the research identified a gene SPAG5 which, was the most influential in the proliferation system and indicates which patients will benefit from chemotherapy, meaning SPAG5 could be a new therapeutic target and a biomarker for the tailoring of breast cancer treatment. The study identified a total of 34 biomarkers in the breast cancer proliferation system.

These markers had a calculated probability of false discovery of less than 1x10-73, further demonstrating the efficacy of the ANNi approach. The ANNi methodology was also successful in specifying influential genes that could be new target markers for childhood sarcomas (R4); in identifying, via a commercial contract with Syngenta, transcriptomic regulators of ripening in the tomato (R5); and in modelling the ATR-CHEK1 system (critical for genomic stability) to predict and validate therapeutic targets for breast cancer. Ball’s ANN-based technologies have been applied to numerous clinical questions and the results described in 200-plus publications.

Evidence

NTU's spin-out company Intelligent OMICS Ltd. has commercialised NTU IP and generated significant revenue and finance investment

Successful business Intelligent OMICS Ltd has been built through the commercial exploitation of the ANN methods and ANNi machine learning algorithms developed at NTU for in silico biomarker and drug target discovery. The company was rebranded as Intelligent OMICS Ltd in 2019 to better reflect the key technological breakthrough that its Intuitive Informed Intelligence (I3) machine learning platform, built around the NTU research, could now process 50 million models to a given dataset per hour. This innovation has shortened analysis, that would have previously taken six to seven months, down to two to three weeks.

NTU and Intelligent OMICS' research has stimulated foreign direct investment of £4.5m from United States DTRA for different organisations led by the Defence Science and Technology Laboratory

Defence Science and Technology Laboratory (Dstl, an executive agency, sponsored by the UK Ministry of Defence), its military stakeholders, and NTU/Intelligent OMICS Ltd. have received contracts from the US Defence Threat Reduction Agency (DTRA) based on the application of NTU and Intelligent OMICS Ltd. machine learning algorithms. The arising IP is being jointly exploited for a civilian context by Ploughshare Innovations, Dstl’s commercial arm, and Intelligent OMICS.

Commercial adoption of NTU spin-out Intelligent OMICS' algorithms have provided Cumulus Oncology with a significantly more cost-effective method to target drugs for development, and helped to facilitate their successful £1.7m fundraise

Using public and private molecular data Ball through Intelligent OMICS has brought about change in practice and generated commercial income through a pathway-based data mining approach based on Artificial Neural Network Inference ANNi algorithmic process that identifies disease/pathway-specific molecular drivers. This approach replaces the expensive and time-consuming drug screening process with the modelling of biological pathways in a disease. These highly influential molecules operating in a pathway make excellent biological targets for drugs. As a result, Intelligent OMICS has entered into a strategic partnership with Cumulus Oncology. Work with Cumulus has provided insights into key pathways of interest to the company around DNA repair (DDR and CHEK1) in lung cancer. This has enabled Cumulus to define further druggable targets of interest and develop biomarker hypotheses for the drug assets under development, facilitating a recent investment into Cumulus from Private Investors and Scottish Enterprise.

Commercial adoption of Intelligent OMICS' algorithms are being used in Mologic's new, low-cost, rapid diagnostic tests

The ANN algorithms developed by Ball have been offered as a service to diagnostics companies, where they are used to rapidly discover optimised panels of biomarkers, incorporate the biomarker panel into a diagnostic model and deliver that model as a piece of software for incorporation onto a diagnostic device.

Collaboration between Intelligent OMICS and Wuhan Pulmonary Hospital (China) has led to a trial of a new clinical test for latent tuberculosis

The trial has successfully detected latent cases of the disease. This enabled patients to be treated, and also led to a phase 2A trial which has since been completed.

Related staff

Publications

  • R1.G . Ball, S . Mian, F . Holding, R.O. Allibone, J. Lowe, S. Ali, G. Li, S. McCardle, I.O. Ellis, C. Creaser,R.C. Rees. An integrated approach utilising artificial neural networks and SELDI mass spectrometry for the classification of human tumours and rapid identification of potential biomarkers. Bioinformatics 18 (3), pp. 395-404(2002) http://doi.org/10.1093/bioinformatics/18.3.395R2.
  • L.J.Lancashire, D.G.Powe, J.S.Reis-Filho, E. Rakha, C.Lemetre, B.Weigelt, T.M.Abdel-Fatah, A.R.Green, R.Mukta, R.Blamey, E.C.Paish, R.C.Rees, I.O.Ellis, G.R.Ball.A validated gene expression profile for detecting clinical outcome in breast cancer using artificial neural networks. Breast Cancer Research and Treatment120, pp. 83-93 (2010) http://doi.org/10.1007/s10549-009-0378-1R3.
  • T.M.A.Abdel-Fatah, D.Agarwal, D.-X.Liu, R.Russell, O.M.Rueda, K.Liu, B.Xu, P.M.Moseley, A.R.Green, A.G.Pockley, R.C.Rees, C.Caldas, I.O.Ellis, G.R.Ball, S.Y.T.Chan.SPAG5 as a prognostic biomarker and chemotherapy sensitivity predictor in breast cancer: a retrospective, integrated genomic, transcriptomic, and protein analysis. The Lancet Oncology 17(7), pp. 1004-1018 (2016). http://doi.org/10.1016/S1470-2045(16)00174-1R4.
  • D.L.Tong, D.J.Boocock, G.K.R.Dhondalay, C.Lemetre, G.R.Ball.Artificial Neural Network Inference (ANNI): A Study on Gene-Gene Interaction for Biomarkers in Childhood Sarcomas. PLoS ONE9(7), Art. No. e102483 (2014). http://doi.org/10.1371/journal.pone.0102483R5.Y.
  • Pan, G.Bradley, K.Pyke, G.R.Ball, C.Lu, R.Fray, A.Marshall, S.Jayasuta, C.Baxter, R.van Wijk, L.Boyden, R.Cade, N.H.Chapman, P.Faser, C.Hodgman, G.B.Seymour.Network Inference Analysis Identifies an APRR2-Like Gene Linked to Pigment Accumulation in Tomato and Pepper Fruits. Plant physiology161(3), pp/ 1476-1485 (2013) http://doi.org/10.1104/pp.112.212654R6.
  • T.M.A. Abdel-Fatah,F.K. Middleton,A.Arora,D.Agarwal,T. Chen, P.M. Moseley, C.Perry,R.Doherty,S.Chan,A.R. Green,E. Rakha, G.R. Ball, I.O. Ellis, N.J. Curtin, S.Madhusudan. Untangling the ATR-CHEK1 network for prognostication, prediction and therapeutic target validation in breast cancer. Molecular Oncology9(3), pp. 569-585 (2015) http://doi.org/10.1016/j.molonc.2014.10.013