Currently, #variants with higher transmissibility and/or immune evasion (antibody escape) have supplanted the founder strain Wu-Hu-1.

While antibodies from vaccinations and infections induce highly neutralising antibodies against the RBD, the RBD has proven to be incredibly versatile, as shown by the pioneering work of @jbloom_lab @tylernstarr @AllieGreaney in their landmark papers @CELL and @ScienceMagazine.

Running a method called deep mutational scanning (DMS) on the entire RBD (201 amino acids), they were able to determine the impact of single-position substitutions on binding to ACE2, mAbs and serum antibodies.

However, widely circulating variants such as beta as well as newly emerging variants (mu, lambda) possess multiple mutations in the RBD, limiting the utility of single substitution data as it is unable to capture the impact of cooperative mutations.

Sadly, screening the combinatorial sequence space is not an easy feat, as it grows exponentially with increasing number of mutations, rapidly outpacing our screening capabilities.

Even when focusing on just the 20 amino acids directly involved in ACE2 binding, the theoretical sequence space tops out at 1x10^26, far exceeding what can be screened by display technologies, such as Yeast-display. This is where #ML comes into play.

Machine learning tools have recently lead to major breakthroughs. We combat the limits of experimental screening approaches by pairing smart library design with sequence-based ML models.

Our ML models can generalise past the experimental data and interrogate the impact of mutations that haven't been screened physically.

Here, we establish deep mutational learning (#DML), integrating experimental yeast display screening of RBD mutagenesis libraries with deep sequencing and machine learning.

DML provides a comprehensive interrogation of combinatorial RBD mutations and their impact on ACE2 binding as well as immune escape, thus enabling predictive profiling of SARS-COV-2 variants.

Through physically screening the RBD by yeast display, RBD variants all the way up to 7 mutations away from Wu-Hu-1 were predicted for ACE2 binding (92% accuracy) and antibody escape (94% accuracy).

#DML could be used as a de facto #monitoring system by rapidly and efficiently making predictions on #genomic #surveillance data of new variants, without the immediate need for experimental assays.

Given the emergence of highly mutated variants such as #Omicron, crucial public health decisions have to be made swiftly, often before experimental assays can be performed.

#DML may enable researchers to select candidate antibody therapeutics and cocktails with the broadest efficacy against the spectrum of possible variants, some of which may occur simultaneously and may be highly mutated such as #Omicron. Do we want to be proactive or reactive?

Applying #DML to predict #SARSCoV2 escape from serum of vaccinated or convalescent individuals, combined with phylogenetic models of #viral #evolution may enable the prospective identification of future emerging variants.

*TL;DR* ML models trained on thousands of classified RBD variants obtained from library screening make highly accurate predictions across a sequence space of billions of RBD variants, several orders of magnitude larger than what is possible from experimental screening alone.

Great collaborative effort of PostDoc and PhD students from our lab @JosephMTaft @cedricrweber @beichengao @Roy_ehling @AlexYermanos as well as friends from @waikato Dr William Kelton @saireddy911

Update: Our deep mutational learning #DML paper on #SARSCoV2 profiling is now online at @CellCellPress! Now with more in-depth assessments of a greater number of antibodies! https://t.co/SIh8ltLl2E

@ReddyLab_ETHZ @saireddy911 @CellCellPress Congratulations Sai!