A Deep-Learning Sequence-Based Method to Predict Protein Stability Changes Upon Genetic Variations

Abstract

Several studies have linked disruptions of protein stability and its normal functions to disease. Therefore, during the last few decades, many tools have been developed to predict the free energy changes upon protein residue variations. Most of these methods require both sequence and structure information to obtain reliable predictions. However, the lower number of protein structures available with respect to their sequences, due to experimental issues, drastically limits the application of these tools. In addition, current methodologies ignore the antisymmetric property characterizing the thermodynamics of the protein stability: a variation from wild-type to a mutated form of the protein structure (XW → XM) and its reverse process (XM → XW) must have opposite values of the free energy difference (ΔΔGWM = – ΔΔGMW). Here we propose ACDC-NN-Seq, a deep neural network system that exploits the sequence information and is able to incorporate into its architecture the antisymmetry property. To our knowledge, this is the first convolutional neural network} to predict protein stability changes relying solely on the protein sequence. We show that ACDC-NN-Seq compares favorably with the existing sequence-based methods.

Predicting protein stability changes upon single-point mutation: a thorough comparison of the available tools on a new dataset

“…ACDC-NN [ 31 ] and its sequence-based version ACDC-NN-Seq [ 32 ] (stand-alone tool): neural network-based methods whose architectures satisfy the antisymmetry properties by construction. They both take as input the local information from the amino acids in the neighbourhood of the mutation and they both use multiple sequence alignments considering the two amino acids involved in the mutation.…”

PROST: AlphaFold2-aware Sequence-Based Predictor to Estimate Protein Stability Changes upon Missense Mutations

“…The benchmarked predictors included sequence-based and structure-based predictors for a fair comparison. Sequence-based predictors included IMutant, iStable, STRUM, EASE-MM, INPS, BoostDDG, iStable2.0_SEQ, SAAFEC-SEQ, DDGun, and ACDC-NN-Seq . On the basis of the current literature, it was concluded that sequence-based predictors might have performance issues, so we also included the structure-based predictors, including SDM, FoldX, mCSM, DUET, MAESTRO, PoPMuSiC, SDM2, DeepDDG, iDeepDDG, PoPMuSiCSym, and DDGun3D for the performance benchmarking.…”

“…Sequence-based predictors included IMutant, 45 iStable, 17 STRUM, 46 EASE-MM, 47 INPS, 18 BoostDDG, 20 iStable2.0_SEQ, 48 SAAFEC-SEQ, 21 DDGun, 12 and ACDC-NN-Seq. 19 On the basis of the current literature, it was concluded that sequence-based predictors might have performance issues, 3 so we also included the structure-based predictors, including SDM, 9 FoldX, 8 mCSM, 13 DUET, 15 MAESTRO, 16 PoPMuSiC, 49 SDM2, 10 DeepDDG, 14 iDeepDDG, 14 PoPMuSiCSym, 11 and DDGun3D 12 for the performance benchmarking. Here, we performed multiple cross-validation tests using the S2647 direct mutations, S2647 inverse mutations, and S5294 (S2647 direct mutations + S2647 inverse mutations) training data sets.…”

“…Usually, structure-based predictors use information on the structural environment of the mutated residue as well as the statistical and physical potentials to achieve a high predictive performance. A small number of predictors, such as iStable, 17 INPS, 18 the sequence-based version of iStable2.0 and DDGun, 12 ACDC-NN-Seq, 19 EASE-MM, BoostDDG, 20 and SAAFEC-SEQ, 21 use sequence-based information; they are employed when no protein structure is available. 18 Recently, we critically reviewed structure-and sequencebased predictors reported over the past decade 3 and discussed their strengths and limitations.…”

“…Indeed, some recent work indicates that some sequence-based predictors, such as SAAFEC-SEQ, BoostDDG, and DDGun, have a promising performance when evolutionary-based features of protein sequences are utilized. Also, another latest work called ACDC-NN-Seq shows outstanding performance in comparison to state-of-the-art predictors; ACDC-NN-Seq uses a deep neural network on sequence information and incorporates both direct and inverse mutations for training/testing.…”