Experimental Solvation Free Energies Research Articles

Force field for halide and alkali ions in water based on single-ion and ion-pair thermodynamic properties for a wide range of concentrations.

A classical non-polarizable force field for the common halide (F-, Cl-, Br-, and I-) and alkali (Li+, Na+, K+, and Cs+) ions in SPC/E water is presented. This is an extension of the force field developed by Loche et al. for Na+, K+, Cl-, and Br- (JPCB 125, 8581-8587, 2021): in the present work, we additionally optimize Lennard-Jones parameters for Li+, I-, Cs+, and F- ions. Li+ and F- are particularly challenging ions to model due to their small size. The force field is optimized with respect to experimental solvation free energies and activity coefficients, which are the necessary and sufficient quantities to accurately reproduce the electrolyte thermodynamics. Good agreement with experimental reference data is achieved for a wide range of concentrations (up to 4 mol/l). We find that standard Lorentz-Berthelot combination rules are sufficient for all ions except F-, for which modified combination rules are necessary. With the optimized parameters, we show that, although the force field is only optimized based on thermodynamic properties, structural properties are reproduced quantitatively, while ion diffusion coefficients are in qualitative agreement with experimental values.

Predicting Solvation Free Energies from the Minnesota Solvation Database Using Classical Density Functional Theory Based on the PC-SAFT Equation of State.

We critically assess the capabilities of classical density functional theory (DFT) based on the perturbed-chain statistical associating fluid theory (PC-SAFT) equation of state to predict the solvation free energies of small molecules in various hydrocarbon solvents. We compare DFT results with experimental data from the Minnesota solvation database and utilize statistical methods to analyze the accuracy of our approach, as well as its weaknesses. The mean absolute error of the solvation free energies is 3.7 kJ mol-1 for n-alkane solvents, ranging from pentane to hexadecane, with 473 solute-solvent systems. For solvents consisting of cyclic hydrocarbons (cyclohexane, benzene, toluene, and ethylbenzene) with 245 solute-solvent systems, we report a slightly larger mean absolute error of 4.2 kJ mol-1. We identify three possible sources of errors: (i) the neglect of solute-solvent and solvent-solvent Coulomb interactions, which limits the applicability of PC-SAFT DFT to nonpolar and weakly polar molecules; (ii) the solute's Lennard-Jones parameters supplied by the general AMBER force field, which are not parametrized toward solvation free energies; and (iii) the application of the Lorentz-Berthelot combining rules to the dispersive interactions between a segment of the PC-SAFT solvent and a Lennard-Jones interaction site of the solute. The approach is more accurate than standard implementations of phenomenological models in common chemistry software packages, which exhibit mean absolute errors larger than 9.12 kJ mol-1, even though newer phenomenological models achieve a mean absolute error of about 2 kJ mol-1. PC-SAFT DFT is more computationally efficient than state of the art explicit molecular simulations in combination with free energy perturbation methods. It is predictive with respect to solvation free energies, i.e., the input for the model is the (element-specific) molecular force field, the solute configuration from molecular dynamics simulations, and the (substance-specific) PC-SAFT parameters. The PC-SAFT parametrization uses pure-component data and does not require experimental solvation free energies. The PC-SAFT equation of state, without applying a DFT formalism, can also be used to calculate solvation free energies, provided that the PC-SAFT parameters for the solute are available. A large number of substances was recently parametrized by members of our group (Esper, T.; Bauer, G.; Rehner, P.; Gross, J. Ind. Eng. Chem. Res. 2023, 62), which enables a comparison to the DFT approach for 103 substances. Accurate results are obtained from the PC-SAFT equation of state with an MAE below 2.51 kJ mol-1. The DFT approach does not require PC-SAFT parameters for the solute and can be applied to all solutes that can be represented by the molecular force field.

Development and test of highly accurate endpoint free energy methods. 3: partition coefficient prediction using a Poisson-Boltzmann method combined with a solvent accessible surface area model for SAMPL challenges.

Accurately predicting solvation free energy is the key to predict protein-ligand binding free energy. In addition, the partition coefficient (log P), which is an important physicochemical property that determines the distribution of a drug in vivo, can be derived directly from transfer free energies, i.e., the difference between solvation free energies (SFEs) in different solvents. Within the Statistical Assessment of the Modeling of Proteins and Ligands (SAMPL) 9 challenge, we applied the Poisson-Boltzmann (PB) surface area (SA) approach to predict the toluene/water transfer free energy and partition coefficient (log Ptoluene/water) from SFEs. For each solute, only a single conformation automatically generated by the free software Open Babel was used. The PB calculation directly adopts our previously optimized boundary definition - a set of general AMBER force field 2 (GAFF2) atom-type based sphere radii for solute atoms. For the non-polar SA model, we newly developed the solvent-related molecular surface tension parameters γ and offset b for toluene and cyclohexane targeting experimental SFEs. This approach yielded the highest predictive accuracy in terms of root mean square error (RMSE) of 1.52 kcal mol-1 in transfer free energy for 16 small drug molecules among all 18 submissions in the SAMPL9 blind prediction challenge. The re-evaluation of the challenge set using multi-conformation strategies based on molecular dynamics (MD) simulations further reduces the prediction RMSE to 1.33 kcal mol-1. At the same time, an additional evaluation of our PBSA method on the SAMPL5 cyclohexane/water distribution coefficient (log Dcyclohexane/water) prediction revealed that our model outperformed COSMO-RS, the best submission model with RMSEPBSA = 1.88 versus RMSECOSMO-RS = 2.11 log units. Two external log Ptoluene/water and log Pcyclohexane/water datasets that contain 110 and 87 data points, respectively, are collected for extra validation and provide an in-depth insight into the error source of the PBSA method.

Solv: An Alternative Continuum Model Implementation Based on Fixed Atomic Charges, Scaled Particle Theory, and the Atom-Atom Potential Method.

An alternative continuum model implementation is reported. The electrostatic contribution to the solvation Gibbs free energy utilizes the noniterative conductor-like screening model of Vyboishchikov and Voityuk (DOI: 10.1002/jcc.26531) based on the fixed partial atomic charges. The nonelectrostatic solute-solvent dispersion-repulsion energy is calculated through the Caillet-Claverie atom-atom potential method employing the grid-based approach. The nonelectrostatic cavitation energy is computed within the scaled particle theory (SPT) formalism with the solute hard-sphere radius obtained via the Pierotti-Claverie (PC) scheme, from the solute molecular surface (SPT-S) or volume (SPT-V). The solvent hard-sphere radius is derived through the fitting to the experimental total solvation free energies of 2530 neutral species in 92 solvents. Application of the model to reproduce both the absolute and relative (reaction net) solvation free energies indicates that the SPT-V approach based on the CM5 charges is the best performer. The method is suggested for the solvation free energy calculation in the nonaqueous solvents.

Engineering Transferable Atomic Force Fields: Empirical Optimization of Hydrocarbon Lennard-Jones Interactions by Direct Mapping of Parameter Space.

The utility of atomistic simulations depends on the accuracy of the force field used to represent the potential energy landscape, the consistency with which interaction parameters can be assigned, and the extent to which parameters can be transferred between chemical entities. Here, parameter space mapping, a simple and robust procedure for atom typing (parameter assignment) and parameter optimization, is used to identify a minimal set of parameters capable of simultaneously reproducing the density, heat of vaporization, and solvation free energies for a targeted set of simple hydrocarbons. Using an atom-centered fixed charge model and a 6-12 Lennard-Jones potential, the experimental densities and the heats of vaporization for 22 hydrocarbons (linear, cyclic, and aromatic) could be predicted with high precision: average unsigned error (AUE) of 6.1 kg/m3 and 0.5 kJ/mol, respectively, and R2 values of 0.991 and 0.999, respectively. For the 17 compounds with experimental solvation free energy values in water, the AUE was 1.3 kJ/mol, and the slope and R2 for the line of best fit were 0.968 and 0.991, respectively. A key element in ensuring transferability in this work was minimizing confounding variables by ensuring that the calculation of observables was independent of the precise choice of simulation settings (cutoff, bond constraints, etc.) and the explicit consideration of correlations between parameters.

Development and test of highly accurate endpoint free energy methods. 1: Evaluation of ABCG2 charge model on solvation free energy prediction and optimization of atom radii suitable for more accurate solvationfreeenergy prediction by the PBSA method.

Accurate estimation of solvation free energy (SFE) lays the foundation for accurate prediction of binding free energy. The Poisson-Boltzmann (PB) or generalized Born (GB) combined with surface area (SA) continuum solvation method (PBSA and GBSA) have been widely used in SFE calculations because they can achieve good balance between accuracy and efficiency. However, the accuracy of these methods can be affected by several factors such as the charge models, polar and nonpolar SFE calculation methods and the atom radii used in the calculation. In this work, the performance of the ABCG2 (AM1-BCC-GAFF2) charge model as well as other two charge models, that is, RESP (Restrained Electrostatic Potential) and AM1-BCC (Austin Model 1-bond charge corrections), on the SFE prediction of 544 small molecules in water by PBSA/GBSA was evaluated. In order to improve the performance of the PBSA prediction based on the ABCG2 charge, we further explored the influence of atom radii on the prediction accuracy and yielded a set of atom radius parameters for more accurate SFE prediction using PBSA based on the ABCG2/GAFF2 by reproducing the thermodynamic integration (TI) calculation results. The PB radius parameters of carbon, oxygen, sulfur, phosphorus, chloride, bromide and iodine, were adjusted. New atom types, on, oi, hn1, hn2, hn3, were introduced to further improve the fitting performance. Then, we tuned the parameters in the nonpolar SFE model using the experimental SFE data and the PB calculation results. By adopting the new radius parameters and new nonpolar SFE model, the root mean square error (RMSE) of the SFE calculation for the 544 molecules decreased from 2.38 to 1.05 kcal/mol. Finally, the new radius parameters were applied in the prediction of protein-ligand binding free energies using the MM-PBSA method. For the eight systems tested, we could observe higher correlation between the experiment data and calculation results and smaller prediction errors for the absolute binding free energies, demonstrating that our new radius parameters can improve the free energy calculation using the MM-PBSA method.

The performance of ANI-ML potentials for ligand-n(H2 O) interaction energies and estimation of hydration free energies from end-point MD simulations.

Here, we investigate the performance of "Accurate NeurAl networK engINe for Molecular Energies" (ANI), trained on small organic compounds, on bulk systems including non-covalent interactions and applicability to estimate solvation (hydration) free energies using the interaction between the ligand and explicit solvent (water) from single-step MD simulations. The method is adopted from ANI using the Atomic Simulation Environment (ASE) and predicts the non-covalent interaction energies at the accuracy of wb97x/6-31G(d) level by a simple linear scaling for the conformations sampled by molecular dynamics (MD) simulations of ligand-n(H2 O) systems. For the first time, we test ANI potentials' abilities to reproduce solvation free energies using linear interaction energy (LIE) formulism by modifying the original LIE equation. Our results on ~250 different complexes show that the method can be accurate and have a correlation of R2 = 0.88-0.89 (MAE <1.0kcal/mol) to the experimental solvation free energies, outperforming current end-state methods. Moreover, it is competitive to other conventional free energy methods such as FEP and BAR with 15-20 × fold reduced computational cost.

Understanding the Liquid States of Cyclic Hydrocarbons Containing N, O, and S Atoms via the 3D-RISM-KH Molecular Solvation Theory.

The 3D-reference interaction site model (3D-RISM) molecular solvation theory in combination with the Kovalenko–Hirata (KH) closure is extended to seven heterocyclic liquids to understand their liquid states and to test the performance of the theory in solvation free energy (SFE) calculations of solutes in select solvents. The computed solvent site distribution profiles were compared with the all-atom molecular dynamics (MD) simulations, showing comparable performances. The computational results were compared against the structural parameters for liquids, whenever available, as well as against the experimental SFEs. The liquids are found to have local ordered structures held together via weak interactions in both the RISM and MD simulations. The 3D-RISM-KH computed SFEs are in good agreement with the benchmark values for the tetrahydrothiophene-S,S-dioxide, and showed comparatively larger deviations in the case of the SFEs in the tetrahydrofuran continuum.

Thermodynamics of π-π Interactions of Benzene and Phenol in Water.

The π–π interaction is a major driving force that stabilizes protein assemblies during protein folding. Recent studies have additionally demonstrated its involvement in the liquid–liquid phase separation (LLPS) of intrinsically disordered proteins (IDPs). As the participating residues in IDPs are exposed to water, π–π interactions for LLPS must be modeled in water, as opposed to the interactions that are often established at the hydrophobic domains of folded proteins. Thus, we investigated the association of free energies of benzene and phenol dimers in water by integrating van der Waals (vdW)-corrected density functional theory (DFT) and DFT in classical explicit solvents (DFT-CES). By comparing the vdW-corrected DFT and DFT-CES results with high-level wavefunction calculations and experimental solvation free energies, respectively, we established the quantitative credibility of these approaches, enabling a reliable prediction of the benzene and phenol dimer association free energies in water. We discovered that solvation influences dimer association free energies, but not significantly when no direct hydrogen-bond-type interaction exists between two monomeric units, which can be explained by the enthalpy–entropy compensation. Our comprehensive computational study of the solvation effect on π–π interactions in water could help us understand the molecular-level driving mechanism underlying the IDP phase behaviors.

Accurate Prediction of Aqueous Free Solvation Energies Using 3D Atomic Feature-Based Graph Neural Network with Transfer Learning.

Graph neural network (GNN)-based deep learning (DL) models have been widely implemented to predict the experimental aqueous solvation free energy, while its prediction accuracy has reached a plateau partly due to the scarcity of available experimental data. In order to tackle this challenge, we first build a large and diverse calculated data set Frag20-Aqsol-100K of aqueous solvation free energy with reasonable computational cost and accuracy via electronic structure calculations with continuum solvent models. Then, we develop a novel 3D atomic feature-based GNN model with the principal neighborhood aggregation (PNAConv) and demonstrate that 3D atomic features obtained from molecular mechanics-optimized geometries can significantly improve the learning power of GNN models in predicting calculated solvation free energies. Finally, we employ a transfer learning strategy by pre-training our DL model on Frag20-Aqsol-100K and fine-tuning it on the small experimental data set, and the fine-tuned model A3D-PNAConv-FT achieves the state-of-the-art prediction on the FreeSolv data set with a root-mean-squared error of 0.719 kcal/mol and a mean-absolute error of 0.417 kcal/mol using random data splits. These results indicate that integrating molecular modeling and DL would be a promising strategy to develop robust prediction models in molecular science. The source code and data are accessible at: https://yzhang.hpc.nyu.edu/IMA.

Magnesium force fields for OPC water with accurate solvation, ion-binding, and water-exchange properties: Successful transfer from SPC/E.

Magnesium plays a vital role in a large variety of biological processes. To model such processes by molecular dynamics simulations, researchers rely on accurate force field parameters for Mg2+ and water. OPC is one of the most promising water models yielding an improved description of biomolecules in water. The aim of this work is to provide force field parameters for Mg2+ that lead to accurate simulation results in combination with OPC water. Using 12 different Mg2+ parameter sets that were previously optimized with different water models, we systematically assess the transferability to OPC based on a large variety of experimental properties. The results show that the Mg2+ parameters for SPC/E are transferable to OPC and closely reproduce the experimental solvation free energy, radius of the first hydration shell, coordination number, activity derivative, and binding affinity toward the phosphate oxygens on RNA. Two optimal parameter sets are presented: MicroMg yields water exchange in OPC on the microsecond timescale in agreement with experiments. NanoMg yields accelerated exchange on the nanosecond timescale and facilitates the direct observation of ion binding events for enhanced sampling purposes.

Piecewise Multipole-Expansion Implicit Solvation for Arbitrarily Shaped Molecular Solutes.

The multipole-expansion (MPE) model is an implicit solvation model used to efficiently incorporate solvent effects in quantum chemistry. Even within the recent direct approach, the multipole basis used in MPE to express the dielectric response still solves the electrostatic problem inefficiently or not at all for solutes larger than approximately ten non-hydrogen atoms. In existing MPE parametrizations, the resulting systematic underestimation of the electrostatic solute-solvent interaction is presently compensated for by a systematic overestimation of nonelectrostatic attractive interactions. Even though the MPE model can thus reproduce experimental free energies of solvation of small molecules remarkably well, the inherent error cancellation makes it hard to assign physical meaning to the individual free-energy terms in the model, raising concerns about transferability. Here we resolve this issue by solving the electrostatic problem piecewise in 3D regions centered around all non-hydrogen nuclei of the solute, ensuring reliable convergence of the multipole series. The resulting method thus allows for a much improved reproduction of the dielectric response of a medium to a solute. Employing a reduced nonelectrostatic model with a single free parameter, in addition to the density isovalue defining the solvation cavity, our method yields free energies of solvation of neutral, anionic, and cationic solutes in water in good agreement with experiment.

Transferable Ion Force Fields in Water from a Simultaneous Optimization of Ion Solvation and Ion-Ion Interaction.

The poor performance of many existing nonpolarizable ion force fields is typically blamed on either the lack of explicit polarizability, the absence of charge transfer, or the use of unreduced Coulomb interactions. However, this analysis disregards the large and mostly unexplored parameter range offered by the Lennard-Jones potential. We use a global optimization procedure to develop water-model-transferable force fields for the ions K+, Na+, Cl–, and Br– in the complete parameter space of all Lennard-Jones interactions using standard mixing rules. No extra-thermodynamic assumption is necessary for the simultaneous optimization of the four ion pairs. After an optimization with respect to the experimental solvation free energy and activity, the force fields reproduce the concentration-dependent density, ionic conductivity, and dielectric constant with high accuracy. The force field is fully transferable between simple point charge/extended and transferable intermolecular potential water models. Our results show that a thermodynamically consistent force field for these ions needs only Lennard-Jones and standard Coulomb interactions.

Fast non-iterative calculation of solvation energies for water and non-aqueous solvents.

We propose an efficient and accurate non-iterative method, dubbed uESE, for calculating solvation free energies. Apart from a COSMO-like electrostatic term, the model takes into account non-electrostatic contributions, which depend on atomic surfaces, induced surface charge densities, and the molecular volume. uESE is tested on 35 polar and 57 non-polar solvents. The calculated and experimental solvation free energies are compared for 2892 systems. The method exhibits an excellent performance, which is superior to major solvation methods. The mean absolute error of predicted solvation energies is found below 1 kcal/mol for neutral solutes and below 3 kcal/mol for ions. The calculated data are almost independent of the quantum-chemical method or/and basis sets employed.

Phenol alkylation under phase transfer catalysis conditions: Insights on the mechanism and kinetics from computations

• Additional phenol improves the partition equilibrium. • The RN4+PhO−(PhOH) complex reacts in the organic phase. • Higher aggregates may be involved for Et4N + and Pr4N+. Experiments reported in the literature involving alkylation of the phenol under phase transfer catalysis conditions support the formation of the RN 4 + PhO ― (PhOH) complex in the organic phase. This complex could be directly involved in the transition state. In the present theoretical study, the phase transfer catalyzed alkylation of phenol was investigated using M06-2X calculations and solvent effect using the SMD model combined with reliable experimental single-ion solvation free energy data in aqueous solution. The analysis was done for tetraethylammonium, tetrapropylammonium and tetrabutylammonium phase transfer catalysts. In the case of tetrabutylammonium ion, the calculations have predicted a reaction kinetics compatible with experimental data. In addition, the BuN 4 + PhO ― (PhOH) complex is calculated to be the active species in the catalysis. For the tetraethylammonium and tetrapropylammonium cations, both the RN 4 + PhO ― and RN 4 + PhO ― (PhOH) complexes are not able to explain the catalysis, suggesting that another more complex aggregate can be involved in the process.

Nonbonded Zr4+ and Hf4+ Models for Simulations of Condensed Phase Metal–Organic Frameworks

Recently, the practical applications of tetravalent metal-based metal–organic frameworks (MOFs) have been drawing increasing attention due to their excellent stability and varied functionality. Understanding the behavior of MOFs at the atomic level is important for their rational design and postsynthetic optimization. In classical simulations, the cationic dummy atom model, where the charge of the metal ion is distributed to surrounding dummy particles, has shown robustness in providing the appropriate coordination geometry while retaining the flexibility in the metal ligand interaction. Here, we present a set of eight-coordinated tetravalent metal (Zr4+, Hf4+) dummy model force fields with 12-6-4 type Lennard–Jones potential for describing ion-induced dipole interactions. This model can reproduce both the experimental solvation free energy of the bare ion and good metal–ligand distances and has been validated as useful and transferable among MOFs with the same metal-core topology. The model is inherently modular since nodes and linkers are independently parametrized. It was tested with three different solvents which all led to stable structures. It also was been further validated with different metal node structures and different linkers, Zr/Hf mixed MOFs, and defects containing MOFs. Since the model is nonbonded, the process of metal–ligand substitutions could be studied. The flexibility of the model makes its applicability to chemical problems broad.

Evaluating Classical Force Fields against Experimental Cross-Solvation Free Energies.

Experimental solvation free energies are nowadays commonly included as target properties in the validation and sometimes even in the calibration of condensed-phase force fields. However, this is often done in a nonsystematic fashion, by considering available solvation free energies involving an arbitrary collection of solutes in a limited set of solvents (e.g., water, octanol, chloroform, cyclohexane, or hexane). Here, this approach is made more systematic by introducing the concept of cross-solvation free energies ΔsGA:B⊖ for a set of N molecules that are all in the liquid state under ambient conditions, namely the matrix of N2 entries for ΔsGA:B⊖ considering each of the N molecules either as a solute (A) or as a solvent (B). Relying on available experimental literature followed by careful data curation, a complete ΔsGA:B⊖ matrix of 625 entries is constructed for 25 molecules with one to seven carbon atoms representative for alkanes, chloroalkanes, ethers, ketones, esters, alcohols, amines, and amides. This matrix is then used to compare the relative accuracies of four popular condensed-phase force fields: GROMOS-2016H66, OPLS-AA, AMBER-GAFF, and CHARMM-CGenFF. In broad terms, and in spite of very different force-field functional-form choices and parametrization strategies, the four force fields are found to perform similarly well. Relative to the experimental values, the root-mean-square errors range between 2.9 and 4.0 kJ·mol-1 (lowest value of 2.9 for GROMOS and OPLS), and the average errors range between -0.8 and +1.0 kJ·mol-1 (lowest magnitude of 0.2 for AMBER and CHARMM). These differences are statistically significant but not very pronounced, especially considering the influence of outliers, some of which possibly caused by inaccurate experimental data.

Comparing Alchemical Free Energy Estimates to Experimental Values Based on the Ben-Naim Formula: How Much Agreement Can We Expect?

The solvation free energy (SFE) plays a key role in thermodynamics. One well-established method for computing the SFE is through an alchemical transformation. However, experimental SFEs are generally determined according to the Ben-Naim equations relying on vapor pressure or density ratios. It is important to establish whether, or to what extent, typical alchemical-based free energy computations provide results comparable to experimental SFEs. In this work, we mimic experimental measurements by simulating the liquid-vapor coexistence of water without alchemical operations. The SFEs measured through vapor pressure and density ratios are used to validate the SFEs obtained through alchemical transformations. It is shown that proper consideration of the nonideal behavior of the vapor is important to ensure that the alchemical SFEs are consistent with the Ben-Naim SFEs. Alchemical transformations in the vapor phase should be performed in addition to solution phase transformations for strongly interacting solutes, such as those with low boiling temperatures and large second virial coefficients. A formula based on the virial expansion of pressure is proposed to provide a better estimate of the true SFE from the simulated vapor pressures. The proposed formula is also applicable to experimental determinations of SFE when the pressure-based route is used.

Hydration of Aromatic Heterocycles as an Adversary of π-Stacking.

Hydration is one of the key players in the protein–ligand binding process. It not only influences the binding process per se, but also the drug’s absorption, distribution, metabolism, and excretion properties. To gain insights into the hydration of aromatic cores, the solvation thermodynamics of 40 aromatic mono- and bicyclic systems, frequently occurring in medicinal chemistry, are investigated. Thermodynamics is analyzed with two different methods: grid inhomogeneous solvation theory (GIST) and thermodynamic integration (TI). Our results agree well with previously published experimental and computational solvation free energies. The influence of adding heteroatoms to aromatic systems and how the position of these heteroatoms impacts the compound’s interactions with water is studied. The solvation free energies of these heteroaromatics are highly correlated to their gas phase interaction energies with benzene: compounds showing a high interaction energy also have a high solvation free energy value. Therefore, replacing a compound with one having a higher gas phase interaction energy might not result in the expected improvement in affinity. The desolvation costs counteract the higher stacking interactions, hence weakening or even inverting the expected gain in binding free energy.

Accurate and transferable multitask prediction of chemical properties with an atoms-in-molecules neural network.

Atomic and molecular properties could be evaluated from the fundamental Schrodinger's equation and therefore represent different modalities of the same quantum phenomena. Here, we present AIMNet, a modular and chemically inspired deep neural network potential. We used AIMNet with multitarget training to learn multiple modalities of the state of the atom in a molecular system. The resulting model shows on several benchmark datasets state-of-the-art accuracy, comparable to the results of orders of magnitude more expensive DFT methods. It can simultaneously predict several atomic and molecular properties without an increase in the computational cost. With AIMNet, we show a new dimension of transferability: the ability to learn new targets using multimodal information from previous training. The model can learn implicit solvation energy (SMD method) using only a fraction of the original training data and an archive median absolute deviation error of 1.1 kcal/mol compared to experimental solvation free energies in the MNSol database.