Search

Machine learning has been proposed as an alternative to theoretical modeling when dealing with complex problems in biological physics. However, in this perspective, we argue that a more successful approach is a proper combination of these two methodologies. We discuss how ideas coming from physical modeling neuronal processing led to early formulations of computational neural networks, e.g., Hopfield networks. We then show how modern learning approaches like Potts models, Boltzmann machines, and the transformer architecture are related to each other, specifically, through a shared energy representation. We summarize recent efforts to establish these connections and provide examples on how each of these formulations integrating physical modeling and machine learning have been successful in tackling recent problems in biomolecular structure, dynamics, function, evolution, and design. Instances include protein structure prediction; improvement in computational complexity and accuracy of molecular dynamics simulations; better inference of the effects of mutations in proteins leading to improved evolutionary modeling and finally how machine learning is revolutionizing protein engineering and design. Going beyond naturally existing protein sequences, a connection to protein design is discussed where synthetic sequences are able to fold to naturally occurring motifs driven by a model rooted in physical principles. We show that this model is "learnable" and propose its future use in the generation of unique sequences that can fold into a target structure. [ABSTRACT FROM AUTHOR]

UnlabelledTwo-component systems (TCS) that employ histidine kinases (HK) and response regulators (RR) are critical mediators of cellular signaling in bacteria. In the model cyanobacterium Synechococcus elongatus PCC 7942, TCSs control global rhythms of transcription that reflect an integration of time information from the circadian clock with a variety of cellular and environmental inputs. The HK CikA and the SasA/RpaA TCS transduce time information from the circadian oscillator to modulate downstream cellular processes. Despite immense progress in understanding of the circadian clock itself, many of the connections between the clock and other cellular signaling systems have remained enigmatic. To narrow the search for additional TCS components that connect to the clock, we utilized direct-coupling analysis (DCA), a statistical analysis of covariant residues among related amino acid sequences, to infer coevolution of new and known clock TCS components. DCA revealed a high degree of interaction specificity between SasA and CikA with RpaA, as expected, but also with the phosphate-responsive response regulator SphR. Coevolutionary analysis also predicted strong specificity between RpaA and a previously undescribed kinase, HK0480 (herein CikB). A knockout of the gene for CikB (cikB) in a sasA cikA null background eliminated the RpaA phosphorylation and RpaA-controlled transcription that is otherwise present in that background and suppressed cell elongation, supporting the notion that CikB is an interactor with RpaA and the clock network. This study demonstrates the power of DCA to identify subnetworks and key interactions in signaling pathways and of combinatorial mutagenesis to explore the phenotypic consequences. Such a combined strategy is broadly applicable to other prokaryotic systems.ImportanceSignaling networks are complex and extensive, comprising multiple integrated pathways that respond to cellular and environmental cues. A TCS interaction model, based on DCA, independently confirmed known interactions and revealed a core set of subnetworks within the larger HK-RR set. We validated high-scoring candidate proteins via combinatorial genetics, demonstrating that DCA can be utilized to reduce the search space of complex protein networks and to infer undiscovered specific interactions for signaling proteins in vivo Significantly, new interactions that link circadian response to cell division and fitness in a light/dark cycle were uncovered. The combined analysis also uncovered a more basic core clock, illustrating the synergy and applicability of a combined computational and genetic approach for investigating prokaryotic signaling networks.

Identification of novel drug targets and chemotherapeutic agents is a high priority in the fight against cancer. Here, we report that MAD-28, a designed cluvenone (CLV) derivative, binds to and destabilizes two members of a unique class of mitochondrial and endoplasmic reticulum (ER) 2Fe-2S proteins, mitoNEET (mNT) and nutrient-deprivation autophagy factor-1 (NAF-1), recently implicated in cancer cell proliferation. Docking analysis of MAD-28 to mNT/NAF-1 revealed that in contrast to CLV, which formed a hydrogen bond network that stabilized the 2Fe-2S clusters of these proteins, MAD-28 broke the coordinative bond between the His ligand and the cluster's Fe of mNT/NAF-1. Analysis of MAD-28 performed with control (Michigan Cancer Foundation; MCF-10A) and malignant (M.D. Anderson-metastatic breast; MDA-MB-231 or MCF-7) human epithelial breast cells revealed that MAD-28 had a high specificity in the selective killing of cancer cells, without any apparent effects on normal breast cells. MAD-28 was found to target the mitochondria of cancer cells and displayed a surprising similarity in its effects to the effects of mNT/NAF-1 shRNA suppression in cancer cells, causing a decrease in respiration and mitochondrial membrane potential, as well as an increase in mitochondrial iron content and glycolysis. As expected, if the NEET proteins are targets of MAD-28, cancer cells with suppressed levels of NAF-1 or mNT were less susceptible to the drug. Taken together, our results suggest that NEET proteins are a novel class of drug targets in the chemotherapeutic treatment of breast cancer, and that MAD-28 can now be used as a template for rational drug design for NEET Fe-S cluster-destabilizing anticancer drugs.

A long-standing problem in molecular biology is the determination of a complete functional conformational landscape of proteins. This includes not only proteins' native structures, but also all their respective functional states, including functionally important intermediates. Here, we reveal a signature of functionally important states in several protein families, using direct coupling analysis, which detects residue pair coevolution of protein sequence composition. This signature is exploited in a protein structure-based model to uncover conformational diversity, including hidden functional configurations. We uncovered, with high resolution (mean ~1.9 Å rmsd for nonapo structures), different functional structural states for medium to large proteins (200-450 aa) belonging to several distinct families. The combination of direct coupling analysis and the structure-based model also predicts several intermediates or hidden states that are of functional importance. This enhanced sampling is broadly applicable and has direct implications in protein structure determination and the design of ligands or drugs to trap intermediate states.

Protein-protein interactions are often mediated by flexible loops that experience conformational dynamics on the microsecond to millisecond time scales. NMR relaxation studies can map these dynamics. However, defining the network of inter-converting conformers that underlie the relaxation data remains generally challenging. Here, we combine NMR relaxation experiments with simulation to visualize networks of inter-converting conformers. We demonstrate our approach with the apo Pin1-WW domain, for which NMR has revealed conformational dynamics of a flexible loop in the millisecond range. We sample and cluster the free energy landscape using Markov State Models (MSM) with major and minor exchange states with high correlation with the NMR relaxation data and low NOE violations. These MSM are hierarchical ensembles of slowly interconverting, metastable macrostates and rapidly interconverting microstates. We found a low population state that consists primarily of holo-like conformations and is a "hub" visited by most pathways between macrostates. These results suggest that conformational equilibria between holo-like and alternative conformers pre-exist in the intrinsic dynamics of apo Pin1-WW. Analysis using MutInf, a mutual information method for quantifying correlated motions, reveals that WW dynamics not only play a role in substrate recognition, but also may help couple the substrate binding site on the WW domain to the one on the catalytic domain. Our work represents an important step towards building networks of inter-converting conformational states and is generally applicable.