Volume 8 Issue 1, January 2024

Humanized versions of antibodies with enhanced stability can be designed via the systematic energy-based ranking of computationally grafted non-human complementarity-determining regions onto thousands of human frameworks.

We developed an at-home microsampling approach that measures thousands of metabolites, lipids and proteins in small volumes of blood. Dense multi-omic sampling generates a ‘molecular movie’ that integrates with data from wearables to reveal new insights into the dynamics of human physiology.

We compared a range of linear and nonlinear models based on how accurately they could describe resting-state functional magnetic resonance imaging and intracranial electroencephalography dynamics in humans. Linear autoregressive models were the most accurate in all cases. Using numerical simulations, we demonstrated that spatiotemporal averaging has a critical and robust role in this linearity.

We show that nonlinear latent factors and structures in neural population activity can be modelled in a manner that allows for flexible dynamical inference, causally, non-causally and in the presence of missing neural observations. Further, the developed neural network model improves the prediction of neural activity, behaviour and latent neural structures.

Dense multi-omics microsampling for the frequent capture and analysis of thousands of molecules in blood alongside physiological information from wearables facilitates the profiling of lifestyle-associated changes in an individual’s health.

A computational method for the systematic grafting of animal complementarity-determining regions onto thousands of human frameworks allows for the design of humanized versions of antibodies with improved stability.

Interpretable machine-learning models can identify clinical-stage monoclonal antibodies with optimal combinations of low off-target binding and low self-association in physiological and antibody-formulation conditions.

A multiple-instance-learning model trained to encode and aggregate either the local sequence contexts or the genomic positions of somatic mutations achieved best-in-class performance in classification and prediction tasks.

Linear mathematical models derived from measurements of local field potentials and of whole-brain blood-oxygen-level-dependent neural activity predict resting-state neural dynamics at least as accurately as nonlinear models.

Nonlinear latent factors and latent structures in the activity of neural populations can be computationally modelled to enable flexible inference and to better predict neural activity and behaviour.