AFsample: improving multimer prediction with AlphaFold using massive sampling

SUMMARY

The AlphaFold2 neural network model has revolutionized structural biology with unprecedented performance. We demonstrate that by stochastically perturbing the neural network by enabling dropout at inference combined with massive sampling, it is possible to improve the quality of the generated models. We generated ∼6000 models per target compared with 25 default for AlphaFold-Multimer, with v1 and v2 multimer network models, with and without templates, and increased the number of recycles within the network. The method was benchmarked in CASP15, and compared with AlphaFold-Multimer v2 it improved the average DockQ from 0.41 to 0.55 using identical input and was ranked at the very top in the protein assembly category when compared with all other groups participating in CASP15. The simplicity of the method should facilitate the adaptation by the field, and the method should be useful for anyone interested in modeling multimeric structures, alternate conformations, or flexible structures.

AVAILABILITY AND IMPLEMENTATION

AFsample is available online at http://wallnerlab.org/AFsample.

Gentle and fast all-atom model refinement to cryo-EM densities via a maximum likelihood approach

Better detectors and automated data collection have generated a flood of high-resolution cryo-EM maps, which in turn has renewed interest in improving methods for determining structure models corresponding to these maps. However, automatically fitting atoms to densities becomes difficult as their resolution increases and the refinement potential has a vast number of local minima. In practice, the problem becomes even more complex when one also wants to achieve a balance between a good fit of atom positions to the map, while also establishing good stereochemistry or allowing protein secondary structure to change during fitting. Here, we present a solution to this challenge using a maximum likelihood approach by formulating the problem as identifying the structure most likely to have produced the observed density map. This allows us to derive new types of smooth refinement potential-based on relative entropy-in combination with a novel adaptive force scaling algorithm to allow balancing of force-field and density-based potentials. In a low-noise scenario, as expected from modern cryo-EM data, the relative-entropy based refinement potential outperforms alternatives, and the adaptive force scaling appears to aid all existing refinement potentials. The method is available as a component in the GROMACS molecular simulation toolkit.

Adaptive frequency-based modeling of whole-brain oscillations: Predicting regional vulnerability and hazardousness rates

Whole-brain computational modeling based on structural connectivity has shown great promise in successfully simulating fMRI BOLD signals with temporal coactivation patterns that are highly similar to empirical functional connectivity patterns during resting state. Importantly, previous studies have shown that spontaneous fluctuations in coactivation patterns of distributed brain regions have an inherent dynamic nature with regard to the frequency spectrum of intrinsic brain oscillations. In this modeling study, we introduced frequency dynamics into a system of coupled oscillators, where each oscillator represents the local mean-field model of a brain region. We first showed that the collective behavior of interacting oscillators reproduces previously shown features of brain dynamics. Second, we examined the effect of simulated lesions in gray matter by applying an in silico perturbation protocol to the brain model. We present a new approach to map the effects of vulnerability in brain networks and introduce a measure of regional hazardousness based on mapping of the degree of divergence in a feature space.

Simulations to benchmark time-varying connectivity methods for fMRI

There is a current interest in quantifying time-varying connectivity (TVC) based on neuroimaging data such as fMRI. Many methods have been proposed, and are being applied, revealing new insight into the brain's dynamics. However, given that the ground truth for TVC in the brain is unknown, many concerns remain regarding the accuracy of proposed estimates. Since there exist many TVC methods it is difficult to assess differences in time-varying connectivity between studies. In this paper, we present tvc_benchmarker, which is a Python package containing four simulations to test TVC methods. Here, we evaluate five different methods that together represent a wide spectrum of current approaches to estimating TVC (sliding window, tapered sliding window, multiplication of temporal derivatives, spatial distance and jackknife correlation). These simulations were designed to test each method's ability to track changes in covariance over time, which is a key property in TVC analysis. We found that all tested methods correlated positively with each other, but there were large differences in the strength of the correlations between methods. To facilitate comparisons with future TVC methods, we propose that the described simulations can act as benchmark tests for evaluation of methods. Using tvc_benchmarker researchers can easily add, compare and submit their own TVC methods to evaluate its performance.

A single Markov-type kinetic model accounting for the macroscopic currents of all human voltage-gated sodium channel isoforms

Modelling ionic channels represents a fundamental step towards developing biologically detailed neuron models. Until recently, the voltage-gated ion channels have been mainly modelled according to the formalism introduced by the seminal works of Hodgkin and Huxley (HH). However, following the continuing achievements in the biophysical and molecular comprehension of these pore-forming transmembrane proteins, the HH formalism turned out to carry limitations and inconsistencies in reproducing the ion-channels electrophysiological behaviour. At the same time, Markov-type kinetic models have been increasingly proven to successfully replicate both the electrophysiological and biophysical features of different ion channels. However, in order to model even the finest non-conducting molecular conformational change, they are often equipped with a considerable number of states and related transitions, which make them computationally heavy and less suitable for implementation in conductance-based neurons and large networks of those. In this purely modelling study we develop a Markov-type kinetic model for all human voltage-gated sodium channels (VGSCs). The model framework is detailed, unifying (i.e., it accounts for all ion-channel isoforms) and computationally efficient (i.e. with a minimal set of states and transitions). The electrophysiological data to be modelled are gathered from previously published studies on whole-cell patch-clamp experiments in mammalian cell lines heterologously expressing the human VGSC subtypes (from NaV1.1 to NaV1.9). By adopting a minimum sequence of states, and using the same state diagram for all the distinct isoforms, the model ensures the lightest computational load when used in neuron models and neural networks of increasing complexity. The transitions between the states are described by original ordinary differential equations, which represent the rate of the state transitions as a function of voltage (i.e., membrane potential). The kinetic model, developed in the NEURON simulation environment, appears to be the simplest and most parsimonious way for a detailed phenomenological description of the human VGSCs electrophysiological behaviour.

Role of DARPP-32 and ARPP-21 in the Emergence of Temporal Constraints on Striatal Calcium and Dopamine Integration

In reward learning, the integration of NMDA-dependent calcium and dopamine by striatal projection neurons leads to potentiation of corticostriatal synapses through CaMKII/PP1 signaling. In order to elicit the CaMKII/PP1-dependent response, the calcium and dopamine inputs should arrive in temporal proximity and must follow a specific (dopamine after calcium) order. However, little is known about the cellular mechanism which enforces these temporal constraints on the signal integration. In this computational study, we propose that these temporal requirements emerge as a result of the coordinated signaling via two striatal phosphoproteins, DARPP-32 and ARPP-21. Specifically, DARPP-32-mediated signaling could implement an input-interval dependent gating function, via transient PP1 inhibition, thus enforcing the requirement for temporal proximity. Furthermore, ARPP-21 signaling could impose the additional input-order requirement of calcium and dopamine, due to its Ca²⁺/calmodulin sequestering property when dopamine arrives first. This highlights the possible role of phosphoproteins in the temporal aspects of striatal signal transduction.

A response towards an environmental stimulus could be reinforced if it elicits a reward. On the subcellular level, the environmental stimulus and the reward signal lead to a transient increase in striatal calcium- and dopamine-signaling, respectively. The integration of calcium and dopamine signals, which is important for reward-learning, could elicit a downstream response only if they are close in time and arrive in correct order (first calcium and then dopamine). This study proposes that the requirement for the input signals to be temporally close and in correct order could emerge due to the coordinated signaling via two striatal phosphoproteins, DARPP-32 and ARPP-21. The DARPP-32 signaling implements an input-interval dependent gating function and ARPP-21 implements an input-order dependent threshold-like function. Thus, a molecular mechanism has been presented here which could explain the emergence of important temporal aspects of subcellular signal integration in reward-learning.

DockQ: A Quality Measure for Protein-Protein Docking Models

The state-of-the-art to assess the structural quality of docking models is currently based on three related yet independent quality measures: Fnat, LRMS, and iRMS as proposed and standardized by CAPRI. These quality measures quantify different aspects of the quality of a particular docking model and need to be viewed together to reveal the true quality, e.g. a model with relatively poor LRMS (>10Å) might still qualify as 'acceptable' with a descent Fnat (>0.50) and iRMS (<3.0Å). This is also the reason why the so called CAPRI criteria for assessing the quality of docking models is defined by applying various ad-hoc cutoffs on these measures to classify a docking model into the four classes: Incorrect, Acceptable, Medium, or High quality. This classification has been useful in CAPRI, but since models are grouped in only four bins it is also rather limiting, making it difficult to rank models, correlate with scoring functions or use it as target function in machine learning algorithms. Here, we present DockQ, a continuous protein-protein docking model quality measure derived by combining Fnat, LRMS, and iRMS to a single score in the range [0, 1] that can be used to assess the quality of protein docking models. By using DockQ on CAPRI models it is possible to almost completely reproduce the original CAPRI classification into Incorrect, Acceptable, Medium and High quality. An average PPV of 94% at 90% Recall demonstrating that there is no need to apply predefined ad-hoc cutoffs to classify docking models. Since DockQ recapitulates the CAPRI classification almost perfectly, it can be viewed as a higher resolution version of the CAPRI classification, making it possible to estimate model quality in a more quantitative way using Z-scores or sum of top ranked models, which has been so valuable for the CASP community. The possibility to directly correlate a quality measure to a scoring function has been crucial for the development of scoring functions for protein structure prediction, and DockQ should be useful in a similar development in the protein docking field. DockQ is available at http://github.com/bjornwallner/DockQ/.