Here are some discoveries that fascinate me this week.

Before we start, below is a play list of a recording of The People United Will Never Be Defeated! 36 Piano Variations, played by Igor Levit. The piece was composed by Frederic Rzewsky to pay a tribute to the struggle of Chilean people against the repressive regime of Augusto Pinochet. I discovered it just last week, but I cannot help listening to it again and again.

Drug discovery

Solving problems that both matter and are computationally solvable

Please see my blog post Spotting problems that both matter and are computationally solvable for some thoughts.

Systems Modelling in Pharma - Problem Solving Workshop

I received via the SIAM (Society of Industrial and Applied Mathematics) mailing list an invitation to register the 2020 Systems Modelling in the Pharmaceutical Industry - Problem Solving Workshop, organized by the Fields Institute. Here is the link.

Th workshop aims to equip graduate students and postdoctoral fellows in mathematics and related disciplines with the tools necessary to solve common modelling problems encountered in the pharmaceutical industry. The workshop is aimed for upper-level undergraduates, graduate students, and postdocs. It was originally planned as a one-week workshop. However, due to the online nature of this year’s half a day virtual workshop, registration is open to anyone to whom this may be of interest.

Neutralising antibodies against SARS-Cov-2

The post Lots of Coronavirus Antibody News in the blog In the pipeline run by Derek Lowe provides a timely update of antibody therapeutic candidates against SARS-Cov-2.

It covers two recent reports (Ju et al. and Shi et al., both from Chinese institutes and both published on Nature) about neutralising antibodies against the virus. The antibodies are mostly monoclonal (namely they consist of many different antibodies), and their efficacy seems to be correlated with their affinity to the receptor-binding domain of the viral spike protein.

An interesting fact I learned is that most therapeutic antibodies are dosed around 10mg/kg (data were acquired by Derek Lowe from this interesting tutorial paper on PK of monoclonal antibodies by Ovacik and Lin). In the reports above the dose in animal testing was 50mg/kg. If we translate that number to a human of 70kg, we would need 3.5 gram purified protein per dose. How much is that? From this paper we learn that, for instance, the injection volumes of trastuzumab (a cancer drug targeting Her2/ERBB) at a dose of 6 mg/kg for a patient with a body weight of 70 kg amount to 20 mL for intravenous administration. If the same concentration is used for our antibody, we would need almost 160mL injection. That is almost half of a normal coffee mug! Apparently we need more specific antibodies, or if not available, definitely a lot of them.

It is therefore not surprising that the post highlights the potential bottleneck of antibody production capacity (see the article published on


mTOR regulates stress-induced mutagenesis

Cipponi et al. reported on Nature their findings that mTOR regulates drug resistance. Non-genotoxic drug selection leads to growth limiting stress, which activates DNA damage response and inhibits mTOR signalling, which lead to less accurate DNA repair. The net effect is an increase of genetic diversity and drug resistance.

Using multiple small molecules that target several targets and genome-wide shRNA screening, the team identified mTOR as a common evolutionary capacitor, namely knock-down of genes positively regulating mTOR activity increases genetic diversity. They found that inhibition of mTOR fosters adaptive mutagenesis by repressing accurate DNA repair.

The final figure (Figure 4) of the paper shows the dynamic change of the fitness landscape of cancer cells under pharmacological pressure. In the first phase of adaptation, which takes about 35-45 weeks (shorter than a month), an equilibrium is gradually established between damages induced by the deleterious effect conferred by initial genomic instability and the beneficial effect conferred by resistant genomic configurations. The equilibrium is reached in the second phase of the response.

Nearly PAM-less CRIPSR-Cas9

The CRISPR-Cas9 system of genome editing requires that a short sequence of 2-6 base pairs, known as the protospacer adjacent motif (PAM), immediately follows the DNA sequence targeted by the Cas9 nuclease.

In bacteria, PAM is used to distinguish self from non-self DNA, preventing the CRISPR locus from being targeted and destroyed by the nuclease. In the context of genome editing, if the target DNA sequence is not followed by the PAM sequence, Cas9 will not bind to or cleave the target DNA sequence. Some disease-causing genes, therefore, are not amenable to CRISPR-Cas9 editing.

Walton et al. (Science, 2020) and Chatterjee et al. (Nature Biotechnology, 2020) reported engineered Cas9 enzyme variants that have less requirement about the PAM sequences. While Walton used structure information of Cas9-DNA binding to perform mutagenesis to relax Cas9 PAM preference, Chatterjee started by comparing orthologues of Cas9 and identified a mutation (threonine, abbreviated as T, uncharged, to lysine, abbreviated as K, positively charged) and a positively charged loop sequence that reduce the stringency of PAM dependency and improve the targeting breadth and efficiency of the Cas9 enzyme.

Figure 2b of Chatterjee et al. is especially interesting, which shows that their variant of Cas9 (Sc++ and HiFi-Sc++) show comparable or reduced off-target potential compared with the original Cas9 enzyme. And the off-target potential is dependent on the target: while editing of VEGFA (Vascular endothelial growth factor A, Entrez Gene ID 7422) with Sc++ induces more than 30 off-target sites with less than 6 mismatches (6 in case of HiFi-Sc++), editing EMX1 (empty spiracles homeobox 1. Gene ID 2016) with HiFi-Sc++ induced only 6 off-target sites (0 in case of HiFi-Sc++).

Both studies provide new tools for genome editing. Off-target effects and editing efficiency remain challenges of using these tools in therapeutic settings.

Chimeric antigen receptor (CAR) macrophages

Chimeric antigen receptor (CAR) T cells have been tested to treat haematologic cancer. Their application to solid tumours, however, have turned out to be challenging.

Klinchinsky et al. (Nature Biotechnology 2020) reported engineered CAR macrophages that resemble a pro-inflammatory phenotype (so-called ‘M1’ macrophage, though the dichotomous classification of macrophages into M1 and M2 types is debated). These CAR-Ms express pro-inflammatory cytokines and chemokines, converted bystander M2 macrophages to M1 types, up-regulated genes involved in antigen presentation, recruited T cells and presented antigens to them, and resisted the effects of immunosuppressive cytokines.

Meta-analysis of single-cell datasets for SARS-CoV-2

See my other blog Meta-analysis of single-cell RNA-sequencing studies for SARS-CoV-2.

Mathematics and machine learning

A mathematical perspective of machine learning

The mathematician Weinan E gave a talk on SIAM Mathematics of Data Science Conference with the title A mathematical perspective of Machine Learning. The recording is available on YouTube.



I summarize my learning notes in another blogpost Learning Logstash Part I. The learning was inspired and encouraged by baddila.

Other programming tricks that I learned

autocmd FileType python map <buffer> <F9> :w<CR>:exec '!python3' shellescape(@%, 1)<CR>
autocmd FileType python imap <buffer> <F9> <esc>:w<CR>:exec '!python3' shellescape(@%, 1)<CR>

Other gems

Have a happy weekend!