Next Generation Drug R&D

Macula X’s Technology will Pave the Way for a Brighter Future with Simulated Clinical Trials

Clinical trials are one of the most important milestones and benchmarks when it comes to drug development. They serve to prove the in vivo efficacy of drugs that have been able to pass through phases of in silico (computational) and in vitro screening. These compounds are the ones that show the most promise and after virtually screening through millions of compounds through processes such as computational protein-ligand docking and QSAR toxicity and solubility modeling.

Here’s the issue though — clinical trials take a long time to complete, and the majority of them aren’t even successful.

“An analysis of clinical-trial data from January 2000 up to April 2019 estimated that only around 12% of drug-development programmes ended in success” — Nature

The main reason for this is that there are so many ways that a clinical trial can go completely wrong. I’ll list a few of these below.

• Some trials go through but then confounding factors don’t allow for the determination of the safety of a particular intervention.
• Others have a flawed study experimental design
• Shortages of money and funding occur
• Entering and transferring data from group to group can accrue small mistakes
• Ensuring that the procedure is carried out both accurately and precisely (e.g. patients take the correct dosage).

All of these factors contribute to the fact that clinical trials take so long and that it’s really hard to get drugs into the market with a cap of 1–2 years — we’re experiencing this right now with the COVID-19 crisis and drugs like Remdesivir are being used with medicinal uncertainty through compassionate use. Vaccines are projected to take a minimum of 1–2 years to develop as well alongside small molecule drugs.

If we look at clinical trials empirically, they’re merely a reflection of scientist’s and researchers inability to understand biological complexity at a molecular level. With so many small SNPs, and variations in concentrations of certain species, from the genomic, to the proteomic, to the metabolomic level, it’s seemingly become near impossible to approximate all of this on a computer — thus it seems necessary to implement live clinical trials in order to develop an understanding of side effects of the drug and whether the drug is efficious in reality as opposed to assessing the binding affinity to a particular protein target in silico.

By understanding all of the pathways that a drug is involved in at a much greater depth — for example pharmacokinetic and pharmacodynamic pathways — we predict small molecule side effects at a much greater level

• Pharmacokinetic pathways — pathways involved in the metabolism of a particular drug
• Pharmacodynamic pathways — pathways in involved in the potency and the therapeutic potential of a drug

By mapping out and understanding all of the pathways that involve a drug molecule and — for protein-based therapeutics — understand the different protein protein interactions — we can begin a new era of medicine, one where the development of treatments and cures doesn’t have to take 5–10 years as opposed to a few months of even weeks.

But how do we develop such a robust understanding of all of the pathways and all of the interactions that take place?

Diving into this question involves diving deeper into the field of multi-omics where scientists are combining different layers of biological data to derive high level complex insights about interactions between the different layers. Some interactions relevant to clinical trials.

• Drug-drug interactions (DDIs)
• Protein-protein interactions (PPIs)

By mapping out these two types of interactions, we can begin to understand how to simulate clinical trials on a much greater level of depth, precision and accuracy.

Recently there, have been countless breakthrough studies at the edge of multi-omics that have been able to predict these interactions between drugs, proteins, and between the two — exactly what we need for clinical trials!

In predicting drug protein interactions, there are two main approaches — the chemogenic approach and the pharmacogenomic approach

• chemically similar drugs are expected to interact with chemically similar proteins
• phenotypically similar drugs are expected to interact with chemically similar proteins

As you can see above, these methods can involve unjustifiable extrapolations from drug to drug based on correlations in distantly related properties like chemical similarity and phenotypic similarity, which aren’t directly related to the molecular action occurring at the cellular and molecular levels.

Recent cutting edge research has been using autoencoders to take transcriptional RNA-seq data and compress the transcriptional data into a compressed latent space consisting of specific “disease modules” with genes for a particular disease all clustered in a similar subspace. These types of studies have been groundbreaking as they’ve shown us relationships and identified disease genes that we haven’t recognized before. And this is just using single cell gene expression data! What happens if we interact protein information as well?

A review of the studies involving deep learning algorithms trying to predict motifs of canonical RNA sequences that bind to RNA-binding proteins shows that encoding structure greatly improves the accuracy of models.

What if we could encode all of this protein data into deep learning models and make more robust models that draw generalizations across multiple biological layers? What if we could encode protein structure in small molecule drug discovery machine learning models to draw correlations between specific binding orientations and quantitative metrics relating to clinical trials such as toxicity and solubility?

Encoding protein structures adds a whole new dimension to our current understanding of deep learning and computational simulations as it relates to drug discovery and clinical trials.