BioByte 025: ML for med chem intuition, co-opting a bacterial protein delivery system, de novo protein binder design for disordered proteins, neuroblastoma cell therapy breakthrough
Welcome to Decoding Bio, a writing collective focused on the latest scientific advancements, news, and people building at the intersection of tech x bio. If you’d like to connect or collaborate, please shoot us a note here or chat with us on Twitter: @ameekapadia @ketanyerneni @morgancheatham @pablolubroth @patricksmalone. Happy decoding!
Hello everyone! Quick comment before diving into this week in bio. We’ve been brainstorming the possibility of building an interactive online community for Decoding Bio to discuss the latest science and industry news. If this is something you are interested in, please reach out and let us know! Onwards.
What we read
Getting Inside the Mind of the Medicinal Chemist with Machine Learning [Pat Walters, Practical Cheminformatics, 2023]
Why it matters: Pat Walters, Chief Data Officer at Relay Therapeutics, tests out one common non-ML and one recent ML-based computational approach for replicating the human intuition of expert medical chemists to assess drug-likeness.
Pat Walters, Chief Data Officer at Relay Therapeutics, is back with a new edition of his must-read blog Practical Cheminformatics on machine learning and medicinal chemistry. Medicinal chemistry involves the design, synthesis, and development of chemical compounds for use as therapeutic agents. While it is a quantitative discipline, there is a large degree of “med chem intuition” required by the human chemist as well. Expert medicinal chemists are able to look at a chemical structure and intuit the “drug-likeness” of a molecule - a qualitative concept for how druglike a substance is with respect to factors like bioavailability.
The field has been trying to replicate this intuition in computational algorithms for years. The most widely used approach is the QED method, which uses a weighted combination of a broad set of molecular features like molecular weight, lipophilicity, and topological polar surface area to rate drug-likeness. In this blog, Pat compared QED to a new ML-based method developed by scientists at Novartis and Microsoft called MolSkill. Check out the post for a more detailed comparison of performance between the 2 models, but overall there was a slight advantage for the ML-based method in predicting drug-likeness. Despite the performance advantage of ML, Pat concludes that he is unlikely to use MolSkill over QED because he likes to use tools that he can understand, and because MolSkill is based on a neural network, he views it as a black box. This highlights an important challenge when building computational systems to augment human cognition in tasks that involve some degree of intuition like medicinal chemistry (or diagnostic medicine): users often want model interpretability, or they don’t trust the system enough to use it.
This biotech could one day make human eggs from scratch. But first, they’re trying to rethink IVF [Megan Molteni, STAT, 2023]
The process of IVF requires a large dose of hormones in order to release eggs to improve the odds of fertilization. This is expensive due to the cost of the hormones and often a very uncomfortable process due to the side effects and potentially medium to long-term health risks. Once the eggs have been released and captured, immature eggs are often discarded or matured via in vitro maturation (IVM), with varying degrees of success.
In order to improve this, Gameto has developed Fertilo. Gameto co-cultures the immature eggs with mini-ovaries, or ovarioids, which are stem cell-derived granulosa cells, to produce a complete chemical cocktail to drive maturation better than “guessing the ingredient list and the exact order and amount in which they get added.”
Early results show that 57% of the eggs developed to the blastocyst stage after fertilization and were chromosomally normal compared to 22% in existing IVM methods. Ben Mol from Monash University cautions that from these surrogate outcomes, this appears to be an improvement on existing IVM technology, but it remains to be seen if this will actually result in more live births”.
From Lab to Living Room: mRNA Therapeutics [Chase Feiger, M.D., Forbes 2023]
This week marked the launch of a new bio-focused column on Forbes: From Lab to Living Room, authored by Dr. Chase Feiger, CEO of Ostro. The series aims to increase biofluency for the masses and features a collaboration with Sketchy Learning, a leading cartoon-based edtech company that is how many medical trainees around the world learn and master microbiology.
In this inaugural piece, the Lab to Living Room team takes on mRNA therapeutics with a delightful sketch (below) and video here. The piece also features interviews with scientific and commercial leaders at Moderna and health policy experts about the promises and challenges of mRNA therapies. We’re excited to see this column evolve!
Digital health pioneer Pear Therapeutics files for bankruptcy [Mario Aguilar, STAT, 2023]
Pear Therapeutics, a pioneer for the digital therapeutics industry, filed for Chapter 11 bankruptcy last Friday. They will lay off 92% of their workforce and begin the process of auctioning off their assets. This ends Pear’s long and, at times, promising journey of nine years, coming two years after it went public via SPAC in 2021. If you’ve been following digital therapeutics—prescription-based software interventions—for quite some time, this news may come as a bit of a surprise. Pear boasts getting the first FDA clearance for a prescription digital therapeutic and since received clearance for two other apps including interventions for substance use disorder, opiate use disorder, and insomnia. Despite studies showing positive results with cohorts of patients using Pear’s apps, the company ultimately struggled to get insurance buy-in. Pear’s unfortunate journey highlights that software is not yet thought of in the same way traditional drugs are and insurance coverage is still far from guaranteed. It’s a sobering reminder that in medicine and pharma, regulatory and payer/provider buy-in often matter most.
Programmable protein delivery with a bacterial contractile injection system [Kreitz et al., Nature, 2023]
Kreitz et al., out of the Zhang lab, have engineered an extracellular contractile injection system (eCIS) to develop a programmable protein delivery device. These systems are employed by bacteria to deliver proteins into eukaryotic cells to modulate host biology.
These syringe-like molecules resemble bacteriophages, yet deliver protein rather than nucleic acid cargo. To explore its programmable delivery capabilities, the authors focused on the Photorhabdus virulence cassette (PVC) subtype. These PVCs are usually used by bacteria that reside in nematodes, which in turn infect insects. The PVCs carry toxins to the insect cells.
The team modified E. coli to produce the PVCs and loaded novel payloads into the PVCs including GFP, Cre and ZFN. They then used AlphaFold2 to predict the structure of the tail fibre of the PVC which was hypothesized to be involved in target recognition and then modified it by replacing its target binding domain with a human target-binding domain from Ad5 and others. This resulted in targeting and delivery of functional cargo to adenocarcinoma cells. The authors also demonstrated in vivo targeting of the PVCs intracranially in mice.
We see major applications of this technology, especially in immune privileged regions. Companies such as NanoSyrinx are already working on this approach and we expect a flurry of companies investigating this nature-derived delivery system further. This paper was released shortly after Zhang Lab spin-out Aera, also focused on targeted delivery.
De novo design of modular peptide-binding proteins by superhelical matching [Wu et al., Nature, 2023]
Why it matters: Intrinsically disordered proteins (proteins with regions that do not adopt well-defined, stable three-dimensional structures) are implicated in cancer and neurodegenerative disease, but are a challenging class of proteins to drug. David Baker’s group has developed a new de novo protein design approach for engineering binders to intrinsically disordered regions of proteins.
The latest paper from the Baker Lab at UW adapted their arsenal of de novo protein design tools for a unique and challenging class of targets - proteins that contain intrinsically disordered regions (IDRs). IDRs are stretches of amino acid sequences that do not adopt a well-defined, stable three-dimensional structure under physiological conditions. In contrast to structured domains, which adopt specific conformations, IDRs can exist in multiple conformations that are often dynamic and flexible. While IDRs exist naturally, they have also been linked to a number of diseases such as cancer and neurodegenerative diseases.
Engineering protein binders to targets that have a dynamic and unpredictable shape is difficult. To solve this challenge, the authors chose to target the protein backbone, which has consistent structure in intrinsically disordered proteins. Proteins composed of repeating units were engineered to bind repeating units in the peptide backbone of interest. By matching superhelical parameters (a quaternary protein structure formed by multiple proteins arranged in a helical/twisted structure) between repeating-protein binder and repeating-protein target peptide (while optimizing hydrogen-bonding and hydrophobic interactions), high affinity and specificity binders were designed.
To extend the method beyond non-repetitive sequences to expand the space of IDRs that could theoretically be targeted, the authors chose the human protein ZFC3H1 as a proof-of-concept. They began with a protein previously designed for one of the targets with repeat structure, kept the backbone constant, and redesigned the remaining sequence space using Monte Carlo sampling. Several engineered binders were experimentally validated.
GD2-CART01 for Relapsed or Refractory High-Risk Neuroblastoma [Del Bufalo et al., NEJM, 2023]
Why it matters: Neuroblastoma, one of the most common cancers in children, is extremely aggressive. These children often undergo a series of high risk and morbid interventions such as multiple chemotherapies, surgery, stem cell transplants, and radiation – and the prognosis remains poor. Here, the authors use CAR-T cells targeting the GD2 ganglioside, and demonstrate remarkable outcomes in patients. Now, there is new and renewed hope for neuroblastoma patients.
GD2 is a glycolipid that is expressed at low levels in tissue while markedly elevated in neuroblastoma (amongst other tumor types). As a tumor-associated antigen, GD2 has been targeted previously with anti-GD2 monoclonals with no evidence of overt toxicity - thus, rendering it a promising target for CAR-T. The GD2-CART01 developed here has a few interesting engineered features: 1) the authors used two additional costimulatory endodomains (4-IBB and CD28) in conjunction with the native zeta chain, and 2) they included a suicide switch, which is an inducible DNA construct encoding caspase 9. The suicide switch is simple: if a patient is given rimiducid (a tacrolimus analog), it causes caspase 9 to dimerize, leading to apoptosis.
These results were remarkable: 17/27 children (63%) had a response to treatment; nine children had a complete response, and eight had a partial response. 74% of patients developed mild cytokine release syndrome, of which only one patient required activation of the suicide gene. Thus, GD2-CAR-T is safe and efficacious in treating high-risk neuroblastoma.
Cynomolgus monkey embryo model captures gastrulation and early pregnancy [Li et al., Cell, 2023]
Why it matters: Synthetic embryos, or embryos grown in a lab from stem cells, have been having a moment in science this year. This paper is the first to grow synthetic embryo-like structures from macaque (monkey model) stem cells and reimplant those embryos into the monkeys. This is important because macaques are one of the closest animal models to humans and this research offers a way to study early developmental biology without the ethical concerns of using actual embryos.
We’ve covered a few papers on transforming stem cells into germ cells this year, mostly in the context of in vitro gametogenesis. Synthetic embryos, the focus of this paper, are a little different. Rather than being used for fertility purposes themselves, they are instead tools for studying developmental biology and signals involved in disease, miscarriages, and birth defects. The authors took stem cells from monkey embryos and cultured them into blastocysts. They were able to generate a few layers of the 3-D structure that forms in early embryogenesis including a yolk-sac looking structure. They then implanted the cultured embryos into the wombs of eight healthy female monkeys, with implantation occuring in three of them. Interestingly, the implantation was enough signal to imitate pregnancy and they saw the levels of progesterone and chorionic gonadotropin (what is detected by pregnancy tests) rise. Simultaneously, other embryos were kept in culture in a dish and observed. After several days of culture, the embryos in culture started showing disorganization and failed to resemble normal embryos. In the monkeys, the embryonic sac disappeared and no fetuses formed. The findings that the embryos were short-lived show that the synthetic embryos are not identical to natural embryos and do not behave in the same way over time. The researchers were able to create blastocysts that mimicked structure, but function was not maintained over time which begs the question of are these ‘blastocysts’ just balls of cells or do they hold organizational and reproductive potential, the answer to which influences how this research is regulated from an ethical standpoint.
What we listened to
In case you missed it
What we liked on Twitter
While Twitter is blocking previews of links on Substack, we will have to modify this section. For now, we’ll add screenshots of the tweet and link the image to the original tweet in case you want to expand. Pictures take up more space so we will limit the number of tweets we add.
If you have suggestions on how to get around this limitation, we’re all ears. Drop us a note by responding to this email or below in the comments! Thanks for adapting with us fam.
Did we miss anything? Would you like to contribute to Decoding Bio by writing a guest post? Drop us a note here or chat with us on Twitter: @ameekapadia @ketanyerneni @morgancheatham @pablolubroth @patricksmalone