To build effective therapeutics, biologists iteratively mutate antibody sequences to improve binding and stability. Proposed mutations can be informed by previous measurements or by learning from large antibody databases to predict only typical antibodies. Unfortunately, the space of typical antibodies is enormous to search, and experiments often fail to find suitable antibodies on a budget. We introduce Clone-informed Bayesian Optimization (CloneBO), a Bayesian optimization procedure that efficiently optimizes antibodies in the lab by teaching a generative model how our immune system optimizes antibodies. Our immune system makes antibodies by iteratively evolving specific portions of their sequences to bind their target strongly and stably, resulting in a set of related, evolving sequences known as a *clonal family*. We train a large language model, CloneLM, on hundreds of thousands of clonal families and use it to design sequences with mutations that are most likely to optimize an antibody within the human immune system. We propose to guide our designs to fit previous measurements with a twisted sequential Monte Carlo procedure. We show that CloneBO optimizes antibodies substantially more efficiently than previous methods in realistic *in silico* experiments and designs stronger and more stable binders in *in vitro* wet lab experiments.

Availability of genome sequence, molecular, and clinical phenotype data for large patient cohorts generated by recent technological advances provides an opportunity to dissect the genetic architecture of complex diseases at system level. However, previous analyses of such data have largely focused on the co-localization of SNPs associated with clinical and expression traits, each identified from genome-wide association studies and expression quantitative trait locus mapping. Thus, their description of the molecular mechanisms behind the SNPs influencing clinical phenotypes was limited to the single gene linked to the co-localized SNP. Here we introduce PerturbNet, a statistical framework for learning gene networks that modulate the influence of genetic variants on phenotypes, using genetic variants as naturally occurring perturbation of a biological system. PerturbNet uses a probabilistic graphical model to directly model the cascade of perturbation from genetic variants to the gene network to the phenotype network along with the networks at each layer of the biological system. PerturbNet learns the entire model by solving a single optimization problem with an efficient algorithm that can analyze human genome-wide data within a few hours. PerturbNet inference procedures extract a detailed description of how the gene network modulates the genetic effects on phenotypes. Using simulated and asthma data, we demonstrate that PerturbNet improves statistical power for detecting disease-linked SNPs and identifies gene networks and network modules mediating the SNP effects on traits, providing deeper insights into the underlying molecular mechanisms.

This paper addresses the problem of scalable optimization for L1-regularized conditional Gaussian graphical models. Conditional Gaussian graphical models generalize the well-known Gaussian graphical models to conditional distributions to model the output network influenced by conditioning input variables. While highly scalable optimization methods exist for sparse Gaussian graphical model estimation, state-of-the-art methods for conditional Gaussian graphical models are not efficient enough and more importantly, fail due to memory constraints for very large problems. In this paper, we propose a new optimization procedure based on a Newton method that efficiently iterates over two sub-problems, leading to drastic improvement in computation time compared to the previous methods. We then extend our method to scale to large problems under memory constraints, using block coordinate descent to limit memory usage while achieving fast convergence. Using synthetic and genomic data, we show that our methods can solve problems with millions of variables and tens of billions of parameters to high accuracy on a single machine.

Post-silicon validation has become a crucial part of modern integrated circuit design to capture and eliminate functional bugs that escape pre-silicon verification. The most critical roadblock in post-silicon validation is the limited observability of internal signals of a design, since this aspect hinders the ability to diagnose detected bugs. A solution to address this issue leverage trace buffers: these are register buffers embedded into the design with the goal of recording the value of a small number of state elements, over a time interval, triggered by a user-specified event. Due to the trace buffer's area overhead, only a very small fraction of signals can be traced. Thus, the selection of which signals to trace is of paramount importance in post-silicon debugging and diagnosis. Ideally, we would like to select signals enabling the maximum amount of reconstruction of internal signal values. Several signal selection algorithms for post-silicon debug have been proposed in the literature: they rely on a probability-based state-restoration capacity metric coupled with a greedy algorithm. In this work we propose a more accurate restoration capacity metric, based on simulation information, and present a novel algorithm that overcomes some key shortcomings of previous solutions. We show that our technique provides up to 34% better state restoration compared to all previous techniques while showing a much better trend with increasing trace buffer size.