PI3Kα is a critical oncogenic driver in a broad range of solid tumors, including breast cancer, where activating mutations and pathway hyperactivation are frequently observed. Although two PI3Kα inhibitors have recently been approved by the FDA for breast cancer treatment, their clinical utility is significantly constrained by dose-limiting, on-target toxicities due to wild-type PI3Kα inhibition, most notably insulin resistance and hyperglycemia.

Antibody-drug conjugates (ADCs) offer a compelling strategy to enhance the therapeutic index of PI3Kα-targeted therapies. However, development of such agents has been hindered by the lack of sufficiently potent and selective PI3Kα inhibitors suitable as ADC payloads.

To address this, we utilized Accutar’s proprietary chimeric degrader platform to develop AC4847, a highly potent and selective PI3Kα degrader. By recruiting the Cereblon E3 ligase, AC4847 induced rapid, dose-dependent degradation of PI3Kα and suppression of downstream pAKT signaling across cell lines harboring diverse PI3Kα mutation statuses. AC4847 exhibited sub-nanomolar potency, approximately 100-fold more potent than the small-molecule PI3Kα inhibitor inavolisib, with minimal activity against other PI3K isoforms or Cereblon-associated neosubstrates. When conjugated to TROP2 or HER2 antibodies, AC4847-based degrader-antibody conjugates (DACs) elicited antigen-dependent PI3Kα degradation and robust anti-proliferative effects, with enhanced potency in cell lines expressing high levels of target antigen. In HER2-overexpressing BT-474 and JIMT-1 breast cancer xenograft models, a single dose of trastuzumab-conjugated PI3Kα DAC at 5 mg/kg or lower (administered once every three weeks) achieved durable tumor regression, accompanied by effective PI3Kα degradation and pAKT suppression. Notably, under equivalent plasma DAC exposures, no significant changes in blood glucose or insulin levels were observed in mice, rats, or non-human primates, contrasting with the pronounced metabolic toxicity seen with inavolisib at efficacious exposures.

In conclusion, these preclinical studies demonstrated that PI3Kα-targeted DACs using degrader AC4847 delivered potent, antigen-dependent anti-tumor efficacy while significantly reducing systemic toxicity compared to traditional small-molecule PI3Kα inhibitors. By leveraging the mechanistic advantages of chimeric degraders, such as catalytic target elimination, sustained pathway suppression dependent on protein resynthesis, and the ability to eliminate non-enzymatic scaffold functions, AC4847 offers a novel mode of action that enhances selectivity and durability of response. These attributes, combined with the targeted delivery enabled by antibody conjugation, support the advancement of AC4847-derived DACs into IND-enabling studies as a first-in-class therapeutic strategy for PI3Kα-driven cancers.

Prostate cancer (PCa) is primarily driven by aberrant Androgen Receptor (AR) signaling. Although there has been substantial advancement in antiandrogen therapies, resistance to these treatments remains a significant obstacle, often marked by continuous or enhanced AR signaling in resistant tumors. While the dysregulation of the ubiquitination-based protein degradation process is instrumental in the accumulation of oncogenic proteins, including AR, the molecular mechanism of ubiquitination-driven AR degradation remains largely undefined. We identified UBE2J1 as the critical E2 ubiquitin-conjugating enzyme responsible for guiding AR ubiquitination and eventual degradation. The absence of UBE2J1, found in 5–15% of PCa patients, results in disrupted AR ubiquitination and degradation. This disruption leads to an accumulation of AR proteins, promoting resistance to antiandrogen treatments. By employing a ubiquitination-based AR degrader to adeptly restore AR ubiquitination, we reestablished AR degradation and inhibited the proliferation of antiandrogen-resistant PCa tumors. These findings underscore the fundamental role of UBE2J1 in AR degradation and illuminate an uncharted mechanism through which PCa maintains heightened AR protein levels, fostering resistance to antiandrogen therapies.

We have developed an end-to-end, retrosynthesis system, named ChemiRise, that can propose complete retrosynthesis routes for organic compounds rapidly and reliably. The system was trained on a processed patent database of over 3 million organic reactions. Experimental reactions were atom-mapped, clustered, and extracted into reaction templates. We then trained a graph convolutional neural network-based one-step reaction proposer using template embeddings and developed a guiding algorithm on the directed acyclic graph (DAG) of chemical compounds to find the best candidate to explore. The atom-mapping algorithm and the one-step reaction proposer were benchmarked against previous studies and showed better results. The final product was demonstrated by retrosynthesis routes reviewed and rated by human experts, showing satisfying functionality and a potential productivity boost in real-life use cases.

We propose a method based on neural networks to accurately predict hydration sites in proteins. In our approach, high-quality data of protein structures are used to parametrize our neural network model, which is a differentiable score function that can evaluate an arbitrary position in 3D structures on proteins and predict the nearest water molecule that is not present. The score function is further integrated into our water placement algorithm to generate explicit hydration sites. In experiments on the OppA protein dataset used in previous studies and our selection of protein structures, our method achieves the highest model quality in terms of F1 score, compared to several previous studies.

Molecular modeling is an important topic in drug discovery. Decades of research have led to the development of high quality scalable molecular force fields.

In this paper, we show that neural networks can be used to train a universal approximator for energy potential functions.

By incorporating a fully automated training process we have been able to train smooth, differentiable, and predictive potential functions on large-scale crystal structures. A variety of tests have also been performed to show the superiority and versatility of the machine-learned model.