Weekly AI/ML & Biotech Digest — Sep 14 to Sep 20, 2026

Curated weekly digest of notable AI/ML and biotech developments. Flagged items are manually selected; remaining items are the week's most recent.

🧬 Biotech

• ★ QO‑BRA: A Quantum Operator-Based Autoencoder for De Novo Molecular Design — We introduce a variational quantum autoencoder tailored for de novo molecular design, named QO-BRA (Quantum Operator-Based Real Amplitude autoencoder). QO-BRA leverages quantum circuits for real-amplitude encoding and the SWAP test to estimate reconstruction and latent-space regularization errors during back-propagation. Adjoint encoder and decoder operators enable unitary transformations and a generative process that ensures accurate reconstruction, as well as the novelty, uniqueness, and validity of the generated samples. We showcase the capabilities of QO-BRA as applied to the de novo design of Ca2+-, Mg2+-, and Zn2+-binding metalloproteins after training the generative model with a modest data set.
• ★ The quantum ensemble variational optimization algorithm: Applications to molecular inverse design — Designing molecules with optimized properties remains a fundamental challenge due to the intricate relationship between molecular structure and properties. Traditional computational approaches that address the combinatorial number of possible molecular designs become unfeasible as the molecular size increases, suffering from the so-called “curse of dimensionality” problem. Recent advances in quantum computing hardware present new opportunities to address this problem. Here, we introduce the quantum ensemble variational optimization (QEVO) method for near-term and early fault-tolerant quantum computing platforms. QEVO efficiently maps molecular structures onto an orthonormal basis of binary strings and samples from a superposition state generated by a variational ansatz. The ansatz is iteratively optimized to identify molecular candidates with the desired property. Our numerical simulations demonstrate the potential of QEVO to design drug-like molecules with anticancer properties, operating in combinatorial spaces composed of up to solutions, while employing a shallow quantum circuit that requires only a modest number of qubits. We envision that QEVO could be applied to a wide range of complex problems, offering practical solutions to problems with combinatorial complexity.
• ★ Reimagining research papers as interactive and reliable AI agents
• ★ Evaluating generalization in protein–ligand cofolding methods — Deep learning has driven major breakthroughs in protein structure prediction; however, one of the next critical steps forward is accurately predicting how proteins interact with small-molecule ligands, to enable real-world applications such as drug discovery.
• ★ LLMsFold integrates LLMs and biophysical simulations for de novo drug design — A computational framework combining large language models with biophysical simulations accelerates de novo drug design. (morning-news-biotech, 2026-09-19)
• ★ BiG-Fed: Bilevel Optimization Enhanced Graph-Aided Federated Learning — This paper was pubished 4 years back but I didn't know it until a new connction at the IEEE Quantum week introduced it to me when we discussed federated machine learning. In federated learning, we want to enable differentiated privacy but at the same time prevent a single member brings down the quality of the entire training. How to achieve that is not clear. This paper was recommended as a relevant reading not an answer. Abstract of the paper: Abstract: In federated learning (FL), due to the non-i.i.d. nature of distributedly owned local datasets, personalization is an important design goal. In this paper, we investigate FL scenarios in which data owners are related by a network topology (e.g., traffic prediction based on sensor networks). Existing personalized FL approaches cannot take this information into account. To address this limitation, we propose the Bilevel Optimization enhanced Graph-aided Federated Learning (BiG-Fed) approach. The inner weights enable local tasks to evolve towards personalization, and the outer shared weights on the server side target the non-i.i.d problem enabling individual tasks to evolve towards a global constraint space. To the best of our knowledge, BiG-Fed is the first bilevel optimization technique to enable FL approaches to cope with two nested optimization tasks at the FL server and FL clients simultaneously. Theoretical analysis shows that BiG-Fed is guaranteed to converge in an efficient manner. Extensive experiments on both synthetic and real-world data demonstrate significant superior performance of BiG-Fed over seven state-of-the-art methods.
• ★ GlucoFM: A Dual-Stream Foundation Model for Continuous Glucose Monitoring — Continuous glucose monitoring (CGM) provides a dense view of daily metabolic physiology, yet existing generic time-series and CGM-specific foundation models often encode glucose traces as entangled single-stream sequences, leaving their multiscale temporal structure only implicitly modeled.
• ★ Improving antibody-antigen structure prediction through large-scale distillation of sequence pairs — This a Chinese model for antibody-antigen structure prediction. They claimed better performance in a few tasks compared to AlphaFold3. And the model itself is open source, so likely worth looking into.