The AI-Discovered Aetiology of COVID-19 and Rationale of the Irinotecan+Etoposide Combination Therapy for Critically Ill COVID-19 Patients

Abstract

We present the AI-discovered aetiology of COVID-19, based on a precise disease model of COVID-19 built under five weeks that best matches the epidemiological characteristics, transmission dynamics, clinical features, and biological properties of COVI D-19 and consistently explains the rapidly expanding COVID-19 literature. We present that COVID-19 implements a unique unbiased survival strategy of balancing viral replication with viral spread by increasing its dependence on (i) ACE2-expressing cells for viral entry and spread,(ii) PI3K signaling in ACE2-expressing cells for viral replication and egress, and (iii) viral- non-structural-and-accessory-protein-dependent immunomodulation to balance viral spread and viral replication. We further propose the combination of irinotecan (an in-market topoisomerase I inhibitor) and etoposide (an in-market topoisomerase II inhibitor) could potentially be an exceptionally effective treatment to protect critically ill patients from death caused by COVID-19-specific cytokine storms triggered by sepsis, ARDS, and other fatal comorbidities.

Prospectively validated augmented intelligence for disease-agnostic predictions of clinical success for novel therapeutics

“…The decision-support integration of HAI at the stage of selecting development candidates to enter first-in-human studies could significantly increase the average clinical success rate of investigational new drugs from 7.9% 8 to 90.1% for nearly any disease. 9, 10 This result confirms that exponentially extracted knowledge alone may be sufficient to accurately predict clinical success, effecting a complete reversal of Eroom’s law. 4 The trained HAI is also the world’s first clinically validated model of human aging that could substantially speed up the discovery of preventive medicine for all age-related diseases.…”

“…No clinical data were available worldwide when the preprint of ABNN’s COVID-19 disease model was published on 20. March 2020 24 .…”

“…Humans have designed machine learning methods to train artificial neural networks that have successfully learned black-box models outperforming human experts in playing games 1 and predicting protein structures 2 . Our work shows that machine learning methods can be adapted to train augmented biological neural networks (ABNNs) to successfully learn quasi-white-box models 24 to predict drug clinical efficacy (Figure 2). ABNN could thus safely deliver AI-surpassing clinical utility while averting the numerous challenges in privacy and explainability faced by medical AI systems 21 .…”

“…To evaluate the quality of HAI’s extracted knowledge, we designed the public, prospective prediction of pivotal ongoing clinical trial outcomes at large scale (PROTOCOLS) challenge to benchmark HAI’s real-world performance of predicting drug clinical efficacy without access to human data. HAI achieved a 10.5-fold improvement from the baseline with 99% confidence 7 in the PROTOCOLS validation, readily increasing the average clinical success rate of investigational new drugs from 7.9% 8 to 90% for almost any human diseases 9,10 . The validated HAI confirms that exponentially extracted knowledge alone is sufficient for accurately predicting drug clinical efficacy, effecting a total reversal of Eroom’s law 4 .…”

PI3K/mTOR and topoisomerase inhibitors with potential activity against SARS-CoV-2 infection

“…These findings are consistent with recent studies that propose topoisomerase inhibitors or PI3K/mTOR inhibitors for treatment of COVID-19. One recent study proposed Irinotecan and Etoposide combination therapy for critically ill COVID-19 patients 9 based on the immunomodulatory and viral suppressive profiles of these drugs. Additionally, a recent loss-of-function experiment revealed that certain topoisomerases were required for efficient replication of positive-sense RNA viruses including SARS-CoV-2 10 .…”