Approximating mutual information of high-dimensional variables using learned representations.

Mutual information (MI) is a general measure of statistical dependence with widespread application across the sciences. However, estimating MI between multi-dimensional variables is challenging because the number of samples necessary to converge to an accurate estimate scales unfavorably with dimensionality. In practice, existing techniques can reliably estimate MI in up to tens of dimensions, but fail in higher dimensions, where sufficient sample sizes are infeasible. Here, we explore the idea that underlying low-dimensional structure in high-dimensional data can be exploited to faithfully approximate MI in high-dimensional settings with realistic sample sizes. We develop a method that we call latent MI (LMI) approximation, which applies a nonparametric MI estimator to low-dimensional representations learned by a simple, theoretically-motivated model architecture. Using several benchmarks, we show that unlike existing techniques, LMI can approximate MI well for variables with $> 10^3$ dimensions if their dependence structure has low intrinsic dimensionality. Finally, we showcase LMI on two open problems in biology. First, we approximate MI between protein language model (pLM) representations of interacting proteins, and find that pLMs encode non-trivial information about protein-protein interactions. Second, we quantify cell fate information contained in single-cell RNA-seq (scRNA-seq) measurements of hematopoietic stem cells, and find a sharp transition during neutrophil differentiation when fate information captured by scRNA-seq increases dramatically.

Signal amplification by cyclic extension enables high-sensitivity single-cell mass cytometry.

Mass cytometry uses metal-isotope-tagged antibodies to label targets of interest, which enables simultaneous measurements of ~50 proteins or protein modifications in millions of single cells, but its sensitivity is limited. Here, we present a signal amplification technology, termed Amplification by Cyclic Extension (ACE), implementing thermal-cycling-based DNA in situ concatenation in combination with 3-cyanovinylcarbazole phosphoramidite-based DNA crosslinking to enable signal amplification simultaneously on >30 protein epitopes. We demonstrate the utility of ACE in low-abundance protein quantification with suspension mass cytometry to characterize molecular reprogramming during the epithelial-to-mesenchymal transition as well as the mesenchymal-to-epithelial transition. We show the capability of ACE to quantify the dynamics of signaling network responses in human T lymphocytes. We further present the application of ACE in imaging mass cytometry-based multiparametric tissue imaging to identify tissue compartments and profile spatial aspects related to pathological states in polycystic kidney tissues.

Scalable design of orthogonal DNA barcode libraries.

Orthogonal DNA barcode library design is an essential task in bioengineering. Here we present seqwalk, an efficient method for designing barcode libraries that satisfy a sequence symmetry minimization (SSM) heuristic for orthogonality, with theoretical guarantees of maximal or near-maximal library size under certain design constraints. Seqwalk encodes SSM constraints in a de Bruijn graph representation of sequence space, enabling the application of recent advances in discrete mathematics1 to the problem of orthogonal sequence design. We demonstrate the scalability of seqwalk by designing a library of >106 SSM-satisfying barcode sequences in less than 20 s on a standard laptop.

Multi-micron crisscross structures grown from DNA-origami slats.

Living systems achieve robust self-assembly across a wide range of length scales. In the synthetic realm, nanofabrication strategies such as DNA origami have enabled robust self-assembly of submicron-scale shapes from a multitude of single-stranded components. To achieve greater complexity, subsequent hierarchical joining of origami can be pursued. However, erroneous and missing linkages restrict the number of unique origami that can be practically combined into a single design. Here we extend crisscross polymerization, a strategy previously demonstrated with single-stranded components, to DNA-origami 'slats' for fabrication of custom multi-micron shapes with user-defined nanoscale surface patterning. Using a library of ~2,000 strands that are combinatorially arranged to create unique DNA-origami slats, we realize finite structures composed of >1,000 uniquely addressable slats, with a mass exceeding 5 GDa, lateral dimensions of roughly 2 µm and a multitude of periodic structures. Robust production of target crisscross structures is enabled through strict control over initiation, rapid growth and minimal premature termination, and highly orthogonal binding specificities. Thus crisscross growth provides a route for prototyping and scalable production of structures integrating thousands of unique components (that is, origami slats) that each is sophisticated and molecularly precise.