arXiv subjects

Debora S. Marks

Publications and source records attributed to Debora S. Marks.

11 recordsLinked to original sources

arXiv2025

Few-shot Protein Fitness Prediction via In-context Learning and Test-time Training

Accurately predicting protein fitness with minimal experimental data is a persistent challenge in protein engineering. We introduce PRIMO (PRotein In-context Mutation Oracle), a transformer-based framework that leverages in-context learning and test-time training to adapt rapidly to new proteins and assays without large task-specific datasets. By encoding sequence information, auxiliary zero-shot predictions, and sparse experimental labels from many assays as a unified token set in a pre-training masked-language modeling paradigm, PRIMO learns to prioritize promising variants through a preference-based loss function. Across diverse protein families and properties-including both substitution and indel mutations-PRIMO outperforms zero-shot and fully supervised baselines. This work underscores the power of combining large-scale pre-training with efficient test-time adaptation to tackle challenging protein design tasks where data collection is expensive and label availability is limited.

q-bio.BM↗

arXiv2025

Kernel-Based Evaluation of Conditional Biological Sequence Models

We propose a set of kernel-based tools to evaluate the designs and tune the hyperparameters of conditional sequence models, with a focus on problems in computational biology. The backbone of our tools is a new measure of discrepancy between the true conditional distribution and the model's estimate, called the Augmented Conditional Maximum Mean Discrepancy (ACMMD). Provided that the model can be sampled from, the ACMMD can be estimated unbiasedly from data to quantify absolute model fit, integrated within hypothesis tests, and used to evaluate model reliability. We demonstrate the utility of our approach by analyzing a popular protein design model, ProteinMPNN. We are able to reject the hypothesis that ProteinMPNN fits its data for various protein families, and tune the model's temperature hyperparameter to achieve a better fit.

stat.ML↗

arXiv2024

Guidelines for releasing a variant effect predictor

Computational methods for assessing the likely impacts of mutations, known as variant effect predictors (VEPs), are widely used in the assessment and interpretation of human genetic variation, as well as in other applications like protein engineering. Many different VEPs have been released to date, and there is tremendous variability in their underlying algorithms and outputs, and in the ways in which the methodologies and predictions are shared. This leads to considerable challenges for end users in knowing which VEPs to use and how to use them. Here, to address these issues, we provide guidelines and recommendations for the release of novel VEPs. Emphasising open-source availability, transparent methodologies, clear variant effect score interpretations, standardised scales, accessible predictions, and rigorous training data disclosure, we aim to improve the usability and interpretability of VEPs, and promote their integration into analysis and evaluation pipelines. We also provide a large, categorised list of currently available VEPs, aiming to facilitate the discovery and encourage the usage of novel methods within the scientific community.

q-bio.OT↗

arXiv2022

Tranception: protein fitness prediction with autoregressive transformers and inference-time retrieval

The ability to accurately model the fitness landscape of protein sequences is critical to a wide range of applications, from quantifying the effects of human variants on disease likelihood, to predicting immune-escape mutations in viruses and designing novel biotherapeutic proteins. Deep generative models of protein sequences trained on multiple sequence alignments have been the most successful approaches so far to address these tasks. The performance of these methods is however contingent on the availability of sufficiently deep and diverse alignments for reliable training. Their potential scope is thus limited by the fact many protein families are hard, if not impossible, to align. Large language models trained on massive quantities of non-aligned protein sequences from diverse families address these problems and show potential to eventually bridge the performance gap. We introduce Tranception, a novel transformer architecture leveraging autoregressive predictions and retrieval of homologous sequences at inference to achieve state-of-the-art fitness prediction performance. Given its markedly higher performance on multiple mutants, robustness to shallow alignments and ability to score indels, our approach offers significant gain of scope over existing approaches. To enable more rigorous model testing across a broader range of protein families, we develop ProteinGym -- an extensive set of multiplexed assays of variant effects, substantially increasing both the number and diversity of assays compared to existing benchmarks.

cs.LG↗

arXiv2017

Deep generative models of genetic variation capture mutation effects

The functions of proteins and RNAs are determined by a myriad of interactions between their constituent residues, but most quantitative models of how molecular phenotype depends on genotype must approximate this by simple additive effects. While recent models have relaxed this constraint to also account for pairwise interactions, these approaches do not provide a tractable path towards modeling higher-order dependencies. Here, we show how latent variable models with nonlinear dependencies can be applied to capture beyond-pairwise constraints in biomolecules. We present a new probabilistic model for sequence families, DeepSequence, that can predict the effects of mutations across a variety of deep mutational scanning experiments significantly better than site independent or pairwise models that are based on the same evolutionary data. The model, learned in an unsupervised manner solely from sequence information, is grounded with biologically motivated priors, reveals latent organization of sequence families, and can be used to extrapolate to new parts of sequence space

q-bio.QM↗

arXiv2017

Genetic variation in human drug-related genes

Variability in drug efficacy and adverse effects are observed in clinical practice. While the extent of genetic variability in classical pharmacokinetic genes is rather well understood, the role of genetic variation in drug targets is typically less studied. Based on 60,706 human exomes from the ExAC dataset, we performed an in-depth computational analysis of the prevalence of functional-variants in in 806 drug-related genes, including 628 known drug targets. We find that most genetic variants in these genes are very rare (f < 0.1%) and thus likely not observed in clinical trials. Overall, however, four in five patients are likely to carry a functional-variant in a target for commonly prescribed drugs and many of these might alter drug efficacy. We further computed the likelihood of 1,236 FDA approved drugs to be affected by functional-variants in their targets and show that the patient-risk varies for many drugs with respect to geographic ancestry. A focused analysis of oncological drug targets indicates that the probability of a patient carrying germline variants in oncological drug targets is with 44% high enough to suggest that not only somatic alterations, but also germline variants carried over into the tumor genome should be included in therapeutic decision-making.

q-bio.GN↗

arXiv2016

3D RNA and functional interactions from evolutionary couplings

Non-coding RNAs are ubiquitous, but the discovery of new RNA gene sequences far outpaces research on their structure and functional interactions. We mine the evolutionary sequence record to derive precise information about function and structure of RNAs and RNA-protein complexes. As in protein structure prediction, we use maximum entropy global probability models of sequence co-variation to infer evolutionarily constrained nucleotide-nucleotide interactions within RNA molecules, and nucleotide-amino acid interactions in RNA-protein complexes. The predicted contacts allow all-atom blinded 3D structure prediction at good accuracy for several known RNA structures and RNA-protein complexes. For unknown structures, we predict contacts in 160 non-coding RNA families. Beyond 3D structure prediction, evolutionary couplings help identify important functional interactions, e.g., at switch points in riboswitches and at a complex nucleation site in HIV. Aided by accelerating sequence accumulation, evolutionary coupling analysis can accelerate the discovery of functional interactions and 3D structures involving RNA.

q-bio.BM↗

arXiv2015

Quantification of the effect of mutations using a global probability model of natural sequence variation

Modern biomedicine is challenged to predict the effects of genetic variation. Systematic functional assays of point mutants of proteins have provided valuable empirical information, but vast regions of sequence space remain unexplored. Fortunately, the mutation-selection process of natural evolution has recorded rich information in the diversity of natural protein sequences. Here, building on probabilistic models for correlated amino-acid substitutions that have been successfully applied to determine the three-dimensional structures of proteins, we present a statistical approach for quantifying the contribution of residues and their interactions to protein function, using a statistical energy, the evolutionary Hamiltonian. We find that these probability models predict the experimental effects of mutations with reasonable accuracy for a number of proteins, especially where the selective pressure is similar to the evolutionary pressure on the protein, such as antibiotics.

q-bio.BM↗

arXiv2014

Sequence co-evolution gives 3D contacts and structures of protein complexes

Protein-protein interactions are fundamental to many biological processes. Experimental screens have identified tens of thousands of interactions and structural biology has provided detailed functional insight for select 3D protein complexes. An alternative rich source of information about protein interactions is the evolutionary sequence record. Building on earlier work, we show that analysis of correlated evolutionary sequence changes across proteins identifies residues that are close in space with sufficient accuracy to determine the three-dimensional structure of the protein complexes. We evaluate prediction performance in blinded tests on 76 complexes of known 3D structure, predict protein-protein contacts in 32 complexes of unknown structure, and demonstrate how evolutionary couplings can be used to distinguish between interacting and non-interacting protein pairs in a large complex. With the current growth of sequence databases, we expect that the method can be generalized to genome-wide elucidation of protein-protein interaction networks and used for interaction predictions at residue resolution.

q-bio.BM↗

arXiv2011

3D Protein Structure Predicted from Sequence

The evolutionary trajectory of a protein through sequence space is constrained by function and three-dimensional (3D) structure. Residues in spatial proximity tend to co-evolve, yet attempts to invert the evolutionary record to identify these constraints and use them to computationally fold proteins have so far been unsuccessful. Here, we show that co-variation of residue pairs, observed in a large protein family, provides sufficient information to determine 3D protein structure. Using a data-constrained maximum entropy model of the multiple sequence alignment, we identify pairs of statistically coupled residue positions which are expected to be close in the protein fold, termed contacts inferred from evolutionary information (EICs). To assess the amount of information about the protein fold contained in these coupled pairs, we evaluate the accuracy of predicted 3D structures for proteins of 50-260 residues, from 15 diverse protein families, including a G-protein coupled receptor. These structure predictions are de novo, i.e., they do not use homology modeling or sequence-similar fragments from known structures. The resulting low Cα-RMSD error range of 2.7-5.1Å, over at least 75% of the protein, indicates the potential for predicting essentially correct 3D structures for the thousands of protein families that have no known structure, provided they include a sufficiently large number of divergent sample sequences. With the current enormous growth in sequence information based on new sequencing technology, this opens the door to a comprehensive survey of protein 3D structures, including many not currently accessible to the experimental methods of structural genomics. This advance has potential applications in many biological contexts, such as synthetic biology, identification of functional sites in proteins and interpretation of the functional impact of genetic variants.

q-bio.BM↗

arXiv2011

Direct-coupling analysis of residue co-evolution captures native contacts across many protein families

The similarity in the three-dimensional structures of homologous proteins imposes strong constraints on their sequence variability. It has long been suggested that the resulting correlations among amino acid compositions at different sequence positions can be exploited to infer spatial contacts within the tertiary protein structure. Crucial to this inference is the ability to disentangle direct and indirect correlations, as accomplished by the recently introduced Direct Coupling Analysis (DCA) (Weigt et al. (2009) Proc Natl Acad Sci 106:67). Here we develop a computationally efficient implementation of DCA, which allows us to evaluate the accuracy of contact prediction by DCA for a large number of protein domains, based purely on sequence information. DCA is shown to yield a large number of correctly predicted contacts, recapitulating the global structure of the contact map for the majority of the protein domains examined. Furthermore, our analysis captures clear signals beyond intra- domain residue contacts, arising, e.g., from alternative protein conformations, ligand- mediated residue couplings, and inter-domain interactions in protein oligomers. Our findings suggest that contacts predicted by DCA can be used as a reliable guide to facilitate computational predictions of alternative protein conformations, protein complex formation, and even the de novo prediction of protein domain structures, provided the existence of a large number of homologous sequences which are being rapidly made available due to advances in genome sequencing.

q-bio.QM↗