 Research article
 Open Access
 Published:
MolCycleGAN: a generative model for molecular optimization
Journal of Cheminformatics volume 12, Article number: 2 (2020)
Abstract
Designing a molecule with desired properties is one of the biggest challenges in drug development, as it requires optimization of chemical compound structures with respect to many complex properties. To improve the compound design process, we introduce MolCycleGAN—a CycleGANbased model that generates optimized compounds with high structural similarity to the original ones. Namely, given a molecule our model generates a structurally similar one with an optimized value of the considered property. We evaluate the performance of the model on selected optimization objectives related to structural properties (presence of halogen groups, number of aromatic rings) and to a physicochemical property (penalized logP). In the task of optimization of penalized logP of druglike molecules our model significantly outperforms previous results.
Introduction
The principal goal of the drug design process is to find new chemical compounds that are able to modulate the activity of a given target (typically a protein) in a desired way [1]. However, finding such molecules in the highdimensional chemical space of all molecules without any prior knowledge is nearly impossible. In silico methods have been introduced to leverage the existing chemical, pharmacological and biological knowledge, thus forming a new branch of science—computeraided drug design (CADD) [2, 3]. Computer methods are nowadays applied at every stage of drug design pipelines [2]—from the search of new, potentially active compounds [4], through optimization of their activity and physicochemical profile [5] and simulating their scheme of interaction with the target protein [6], to assisting in planning the synthesis and evaluation of its difficulty [7].
The recent advancements in deep learning have encouraged its application in CADD [8]. The two main approaches are: virtual screening, which uses discriminative models to screen commercial databases and classify molecules as likely active or inactive; de novo design, that uses generative models to propose novel molecules that are likely to possess the desired properties. The former application already proved to give outstanding results [9,10,11,12]. The latter use case is rapidly emerging, e.g. long shortterm memory (LSTM) network architectures have been applied with some success [13,14,15,16].
In the center of our interest are the hittolead and lead optimization phases of the compound design process. Their goals are to optimize the druglike molecules identified in the previous steps in terms of the desired activity profile (increased potency towards given target protein and provision of inactivity towards offtarget proteins) and the physicochemical and pharmacokinetic properties. Optimizing a molecule with respect to multiple properties simultaneously remains a challenge [5]. Nevertheless, some successful approaches to compound generation and optimization have been proposed.
In the domain of molecule generation, Recurrent Neural Networks (RNN) still play a central role. They were successfully applied to SMILES, which is a commonly used text representation of molecules [17, 18]. RNN architectures, especially those based on LSTM or GRU, obtain excellent results in natural language processing tasks where the input is a sequence of tokens that varies in length. Unfortunately, generative models built on SMILES can generate invalid sequences that do not correspond to any molecule. Attempting to solve this problem, grammarbased methods were proposed to ensure the correct contextfree grammar of the output sequence [18,19,20]. Another issue with the SMILES representation is its sensitivity to the structure of the represented molecule. Even small changes in the structural formula of a compound can lead to a very different canonical SMILES, which impacts the ordering of atom processing performed by RNNs. ArúsPous et al. [21] show that randomization of SMILES can substantially improve the quality of generated molecules. Also, several approaches with reinforcement learning at their cores have been used in chemical property optimization [18, 22]. Moreover, RNNs were also successfully applied to molecular graphs, which are in this case constructed node by node [23]. A promising alternative to reinforcement learning is conditional generation, where molecules are generated with the desired properties presented at the input [24, 25].
Variational Autoencoder (VAE) [26] in conjunction with SMILES representation has been used to generate novel molecules from the trained continuous latent space [27, 28]. Additionally, VAE models were also successfully realized directly on molecular graphs [29, 30]. Because of the intermediate continuous representation of the latent space, molecules with similar properties appear in the vicinity of one another. Bayesian optimization can be used to explore this space and find the desired properties [30]. Still, decoding from the latent space is oftentimes nontrivial and requires to determine the ordering of generated atoms when RNNs are used in this process.
Generative Adversarial Networks (GAN) [31] is an alternative architecture that has been applied to de novo drug design. GANs, together with Reinforcement Learning (RL), were recently proposed as models that generate molecules with desired properties while promoting diversity. These models use representations based on SMILES [32, 33], graph adjacency and annotation matrices [34] or are based on graph convolutional policy networks [35]. There are also hybrid approaches which utilize both GANs and latent vector representation in the process of compound generation [36].
To address the problem of generating compounds difficult to synthesize, we introduce MolCycleGAN—a generative model based on CycleGAN [37]—extending the scope of the early version of our method [38] with more advanced experiments and detailed explanations. Given a starting molecule, it generates a structurally similar one but with a desired characteristic. The similarity between these molecules is important for two reasons. First, it leads to an easier synthesis of generated molecules, and second, such optimization of the selected property is less likely to spoil the previously optimized ones, which is important in the context of multiparameter optimization. We show that our model generates molecules that possess desired properties (note that by a molecular property we also mean binding affinity towards a target protein) while retaining their structural similarity to the starting compound. Moreover, thanks to employing graphbased representation instead of SMILES, our algorithm always returns valid compounds.
We evaluate the model’s ability to perform structural transformations and molecular optimization. The former indicates that the model is able to do simple structural modifications such as a change in the presence of halogen groups or number of aromatic rings, and we also consider bioisostere replacement, which is relevant to modern drug optimization process. In the latter, we aim to maximize penalized logP to assess the model’s usefulness for compound design. Penalized logP is chosen because it is a property often selected as a testing ground for molecule optimization models [30, 35], due to its relevance in the drug design process. In the optimization of penalized logP for druglike molecules, our model significantly outperforms previous results. Eventually, experiments on increasing bioactivity are conducted with DRD2 as the biological target. To the best of our knowledge, MolCycleGAN is the first approach to molecule generation that uses the CycleGAN architecture.
Methods
Junction Tree Variational Autoencoder
JTVAE [30] (Junction Tree Variational Autoencoder) is a method based on VAE, which works on graph structures of compounds, in contrast to previous methods which utilize the SMILES representation of molecules [19, 20, 27]. The VAE models used for molecule generation share the encoderdecoder architecture. The encoder is a neural network used to calculate a continuous, highdimensional representation of a molecule in the socalled latent space, whereas the decoder is another neural network used to decode a molecule from coordinates in the latent space. In VAEs the entire encodingdecoding process is stochastic (has a random component). In JTVAE both the encoding and decoding algorithms use two components for representing the molecule: a junctiontree scaffold of molecular subcomponents (called clusters) and a molecular graph [30]. JTVAE shows superior properties compared to SMILESbased VAEs, such as 100\(\%\) validity of generated molecules.
MolCycleGAN
MolCycleGAN is a novel method of performing compound optimization by learning from the sets of molecules with and without the desired molecular property (denoted by the sets X and Y). Our approach is to train a model to perform the transformation \(G: X \rightarrow Y\) and then use this model to perform optimization of molecules. In the context of compound design X and Y can be, e.g., the set of inactive (active) molecules.
To represent the sets X and Y, our approach requires an embedding of molecules which is reversible, i.e. enables both encoding and decoding of molecules.
For this purpose we use the latent space of JTVAE, which is a representation created by the neural network during the training process. This approach has the advantage that the distance between molecules (required to calculate the loss function) can be defined directly in the latent space. Moreover, molecular properties are easier to express on graphs rather than using linear SMILES representation [39]. One could try formulating the CycleGAN model on the SMILES representation directly, but this would raise the problem of defining a differentiable intermolecular distance, as the standard manners of measuring similarity between molecules (Tanimoto similarity) are nondifferentiable.
Our approach extends the CycleGAN framework [37] to molecular embeddings of the latent space of JTVAE [30]. We represent each molecule as a point in the latent space, given by the mean of the variational encoding distribution [26]. Our model works as follows (Fig. 1): (i) we start by defining the sets X and Y (e.g., inactive/active molecules); (ii) we introduce mapping functions \(G: X \rightarrow Y\) and \(F: Y \rightarrow X\); (iii) we introduce discriminator \(D_X\) (and \(D_Y\)) which forces the generator F (and G) to generate samples from a distribution close to the distribution of X (or Y). The components F, G, \(D_X\), and \(D_Y\) are modeled by neural networks (see Workflow for technical details). The main idea of our approach to molecule optimization is to: (i) take the prior molecule x without a specified feature (e.g. specified number of aromatic rings, water solubility, activity) from set X, and compute its latent space embedding; (ii) use the generative neural network G to obtain the embedding of molecule G(x), that has this feature (as if the G(x) molecule came from set Y) but is also similar to the original molecule x; (iii) decode the latent space coordinates given by G(x) to obtain the optimized molecule. Thereby, the method is applicable in lead optimization processes, as the generated compound G(x) remains structurally similar to the input molecule.
To train the MolCycleGAN we use the following loss function:
and aim to solve
We use the adversarial loss introduced in LSGAN [40]:
which ensures that the generator G (and F) generates samples from a distribution close to the distribution of Y (or X), denoted by \(p_{\rm data}^{Y}\) (\(p_{\rm data}^{X}\)).
The cycle consistency loss
reduces the space of possible mapping functions, such that for a molecule x from set X, the GAN cycle brings it back to a molecule similar to x, i.e. F(G(x)) is close to x (and analogously G(F(y)) is close to y). The inclusion of the cyclic component acts as a regularization and may also help in the regime of low data, as the model can learn from both directions of the transformation. This component makes the resulting model more robust (cf. e.g. the comparison [41] of CycleGAN vs noncyclic IcGAN [42]). Finally, to ensure that the generated (optimized) molecule is close to the starting one we use the identity mapping loss [37]
which further reduces the space of possible mapping functions and prevents the model from generating molecules that lay far away from the starting molecule in the latent space of JTVAE.
In all our experiments we use the hyperparameters \(\lambda _1 = 0.3\) and \(\lambda _2 = 0.1\), which were chosen by checking a couple of combinations (for structural tasks) and verifying that our optimization process: (i) improves the studied property and (ii) generates molecules similar to the starting ones. We have not performed a grid search for optimal values of \(\lambda _1\) and \(\lambda _2\), and hence there could be space for improvement. Note that these parameters control the balance between improvement in the optimized property and similarity between the generated and the starting molecule. We show in the Results section that both the improvement and the similarity can be obtained with the proposed model.
Workflow
We conduct experiments to test whether the proposed model is able to generate molecules that possess desired properties and are close to the starting molecules. Namely, we evaluate the model on tasks related to structural modifications, as well as on tasks related to molecule optimization. For testing molecule optimization, we select the octanolwater partition coefficient (logP) penalized by the synthetic accessibility (SA) score and activity towards DRD2 receptor.
logP describes lipophilicity—a parameter influencing a whole set of other characteristics of compounds such as solubility, permeability through biological membranes, ADME (absorption, distribution, metabolism, and excretion) properties, and toxicity. We use the formulation as reported in the paper on JTVAE [30], i.e. for molecule m the penalized logP is given as \(logP(m)SA(m)\). We use the ZINC250K dataset used in similar studies [19, 30], which contains 250 000 druglike molecules extracted from the ZINC database [43].
For DRD2 activity task we use Random Forest classification model trained on ECFP fingerprints as the activity estimator (ROC AUC = 0.92), where the activity data were extracted from the ChEMBL database.
The detailed formulation of the tasks is the following:
Structural transformations: We test the model’s ability to perform simple structural transformations of the molecules. To this end, we choose the sets X and Y, differing in some structural aspects, and then test if our model can learn the transformation rules and apply them to molecules previously unseen by the model. These are the features by which we divide the sets:
Halogen moieties: We split the dataset into two subsets X and Y. The set Y consists of molecules which contain at least one of the following SMARTS: ‘[!#1]Cl’, ‘[!#1]F’, ‘[!#1]I’, ‘C#N’, whereas the set X consists of such molecules which do not contain any of them. The SMARTS chosen in this experiment indicate halogen moieties and the nitrile group. Their presence and position within a molecule can have an immense impact on the compound’s activity.
Bioisosteres: Molecules in set X are molecules with ‘CN’ and without ‘\(\text {CF}_3\)’ group. The set Y consists of molecules which contain ‘\(\text {CF}_3\)’ and does not contain ‘CN’ group.
\({{CF}}_3\) addition: The set X is a random sample from ZINC250K (without ‘\(\text {CF}_3\)’). The set Y consists of molecules which contain ‘\(\text {CF}_3\)’ group. This task is used as a control task for the bioisosteric substitution to check if the model can learn to generate this group at any position.
Aromatic rings: Molecules in X have exactly two aromatic rings, whereas molecules in Y have one or three aromatic rings.
Constrained molecule optimization: We optimize penalized logP, while constraining the degree of deviation from the starting molecule. The similarity between molecules is measured with Tanimoto similarity on Morgan Fingerprints [44]. The sets X and Y are random samples from ZINC250K, where the compounds’ penalized logP values are below and above the median, respectively.
Unconstrained molecule optimization: We perform unconstrained optimization of penalized logP. The set X is a random sample from ZINC250K and the set Y is a random sample from the top20\(\%\) molecules with the highest penalized logP in ZINC250K.
Activity: We use the MolCycleGAN to create active molecules from inactive ones, where DRD2 (dopamine receptor D2) was chosen as the biological target. Compounds with annotated activity towards the target were extracted from ChEMBL database, version 25 [45]. We split the dataset into two subsets, active (Y) and inactive (X). The set Y consists of molecules with \(K_i < 100\) , whereas all remaining molecules are delegated to set X.
Composition of the datasets
Dataset sizes In Tables 1 and 2 we show the number of molecules in the datasets used for training and testing. In all experiments we use separate sets for training the model (\(X_{\text {train}}\) and \(Y_{\text {train}}\)) and separate, nonoverlapping ones for evaluating the model (\(X_{\text {test}}\) and \(Y_{\text {test}}\)). In \(\text {CF}_3\) addition and all physicochemical experiments no \(Y_{\text {test}}\) set is required.
Distribution of the selected properties In the experiment on halogen moieties, the set X always (i.e., both in train and testtime) contains molecules without halogen moieties, and the set Y always contains molecules with halogen moieties. In the dataset used to construct the latent space (ZINC250K) 65% molecules do not contain any halogen moiety, whereas the remaining 35% contain one or more halogen moieties.
In the experiment on aromatic rings, the set X always (i.e., both in train and testtime) contains molecules with 2 rings, and the set Y always contains molecules with 1 or 3 rings. The distribution of the number of aromatic rings in the dataset used to construct the latent space (ZINC250K) is shown in Fig. 2 along with the distribution for X and Y.
In the bioisosteres experiment, the set X always contains molecules with CN group and without \(\text {CF}_3\) group. Set Y always contains molecules with \(\text {CF}_3\) group. In the CF\(_3\) addition experiment, the set X is a random sample from ZINC250K, and the set Y similarly contains molecules with CF\(_3\) group. In the dataset used to construct the latent space (ZINC250K) 5.1% of molecules contain CN group, whereas molecules with \(\text {CF}_3\) group accounts for 3.8% of total dataset.
For the molecule optimization tasks we plot the distribution of the property being optimized (penalized logP) in Fig. 3 (constrained optimization) and Fig. 4 (unconstrained optimization).
In the activity optimization experiment, the set X contains inactive molecules and the set Y contains active molecules. The mean activity prediction equals 0.223 for the whole dataset which was used to construct the latent space (ZINC250K), whereas for the \(X_{\text {test}}\) dataset the mean predicted activity is 0.179.
Architecture of the models
All networks are trained using the Adam optimizer [46] with learning rate 0.0001. During training we use batch normalization [47]. As the activation function we use leakyReLU with \(\alpha = 0.1\). In the structural experiments the models are trained for 100 epochs and in the physicochemical experiments for 300 epochs.
Structural data experiments

Generators are built of one fully connected residual layer, followed by one dense layer. All layers contain 56 units.

Discriminators are built of 6 dense layers of the following sizes: 56, 42, 28, 14, 7, 1 units.
Physicochemical data experiments

Generators are built of four fully connected residual layers. All layers contain 56 units.

Discriminators are built of 7 dense layers of the following sizes: 48, 36, 28, 18, 12, 7, 1 units.
Results and discussion
Structural transformations
In each structural experiment we test the model’s ability to perform simple transformations of molecules in both directions \(X \rightarrow Y\) and \(Y \rightarrow X\). Here, X and Y are nonoverlapping sets of molecules with a specific structural property. We start with experiments on structural properties because they are easier to interpret and the rules related to transforming between X and Y are well defined. Hence, the present task should be easier for the model, as compared to the optimization of complex molecular properties, for which there are no simple rules connecting X and Y.
In Table 3 we show the success rates for the tasks of performing structural transformations of molecules. The task of changing the number of aromatic rings is more difficult than changing the presence of halogen moieties. In the former the transition between X (with 2 rings) and Y (with 1 or 3 rings, cf. Fig. 5) is more than a simple addition/removal transformation, as it is in the other case (see Fig. 5 for the distributions of the aromatic rings). This is reflected in the success rates which are higher for the task of transformations of halogen moieties. In the dataset used to construct the latent space (ZINC250K) 64.9% molecules do not contain any halogen moiety, whereas the remaining 35.1% contain one or more halogen moieties. This imbalance might be the reason for the higher success rate in the task of removing halogen moieties (\(Y \rightarrow F(Y)\)). Molecular similarity and druglikeness are achieved in all experiments.
To confirm that the generated molecules are close to the starting ones, we show in Fig. 6 distributions of their Tanimoto similarities (using Morgan fingerprints). For comparison we also include distributions of the Tanimoto similarities between the starting molecule and a random molecule from the ZINC250K dataset. The high similarities between the generated and the starting molecules show that our procedure is neither a random sampling from the latent space nor a memorization of the manifold in the latent space with the desired value of the property. In Fig. 7 we visualize the molecules, which after transformation are the most similar to the starting molecules.
Bioisosteres
As for the more complicated structural transformation, we present a bioisosteric substitution task. Here, we have sets X and Y with groups CN and CF\(_3\) respectively. These two moieties have similar electronic effects, CN being more hydrophilic. The dataset was constructed so that there are no compounds containing both of these fragments at once. We want to see whether our method can learn to substitute one group with another, or it will put the target group at a random position in the molecule.
Three different optimization procedures are performed: (a) bioisosteric substitution conducted as described above, (b) generating 10 intermediate steps from the bioisosteric substitution optimization path (x, G(x)), and (c) the addition of CF\(_3\) group. In the step wise variant, molecules from the optimization path were taken in equal intervals. In the case of CF\(_3\) addition, we use X without trifluoromethyl group and Y with the group present within the structure. Here, similarly as in the halogen example, we check if our model can learn to include the given substructure in the generated molecule. Here, we treat the CF\(_3\) addition task as a control task for the bioisosteric substitution since it should be easier for the model to add the group in some indefinite position. Figure 8 shows similarities between original and optimized datasets in these three experiments. The plots show that this time the trained transformation leads to more dissimilar molecules, which is probably caused by two major changes in the structure of a compound—first we remove one group, and then we add another group. Comparing similarity distribution to our control task of trifluoromethyl group addition, the latter leads to greater similarity of the generated compounds.
Tables 4 and 5 summarize quantitatively the results of bioisosteric substitution. All the generated molecules maintain high diversity. Interestingly, inverse optimization (substitution of CF\(_3\) group with CN) is an easier task. The reason behind that is probably that CF\(_3\) fragment contains more atoms, and thus its decoding process is more complex. Moreover, it appears that addition of the CF\(_3\) group is a more difficult task than substitution as the success rate is lower here. The higher rates in the substitution variant may be caused by high similarity of two datasets X and Y, which both consist of molecules with one of the two groups with a similar bioactivity effect.
We compare the substituted compounds qualitatively in Figs. 9 and 10. We observe that the moieties are often correctly substituted with only minor changes to the overall compound structure. The method learns to substitute bioisosteric groups rather than attach the new group to other fragments of a molecule. Figure 11 shows the addition scenario, in which again changes to the molecule are small. Additionally, CF\(_3\) group also prefers replacing other atoms, e.g. halogen groups or ketone groups in the examples provided.
Constrained molecule optimization
As our main task we optimize the desired property under the constraint that the similarity between the original and the generated molecule is higher than a fixed threshold (denoted as \(\delta \)). This is a more realistic scenario in drug discovery, where the development of new drugs usually starts with known molecules such as existing drugs [48]. Here, we maximize the penalized logP coefficient and use the Tanimoto similarity with the Morgan fingerprint [44] to define the threshold of similarity, \(sim(m, m') \ge \delta \). We compare our results with previous similar studies [30, 35].
In our optimization procedure each molecule (given by the latent space coordinates x) is fed into the generator to obtain the ‘optimized’ molecule G(x). The pair (x, G(x)) defines what we call an ’optimization path’ in the latent space of JTVAE. To be able to make a comparison with the previous research [30], we start the procedure from the 800 molecules with the lowest values of penalized logP in ZINC250K, and then we decode molecules from \(K = 80\) points along the path from x to G(x) in equal steps.
From the resulting set of molecules we report the molecule with the highest penalized logP score that satisfies the similarity constraint. A modification succeeds if one of the decoded molecules satisfies the constraint and is distinct from the starting one. Figure 12 shows exemplary molecules with highest improvements and high similarity to the starting compounds.
In the task of optimizing penalized logP of druglike molecules, our method significantly outperforms the previous results in the mean improvement of the property (see Table 6). It achieves a comparable mean similarity in the constrained scenario (for \(\delta > 0\)). The success rates are comparable for \(\delta = 0, 0.2\), whereas for the more stringent constraints (\(\delta = 0.4, 0.6\)) our model has lower success rates.
Note that comparably high improvements of penalized logP can be obtained using reinforcement learning [35]. However, many methods using reinforcement learning tend to generate compounds that are not druglike because they suffer from catastrophic forgetting when the optimization task is changed, e.g. they learn the prior druglike distribution first, and then they try to increase the logP property at the cost of divergence from the prior distribution. Nonetheless, this problem can be relatively easily alleviated, e.g., by multitarget optimization that takes QED [49] into account. In our method (as well as in JTVAE) druglikeness is achieved “by design” and is an intrinsic feature of the latent space obtained by training the variational autoencoder on molecules from ZINC (which are druglike).
Molecular paths from constrained optimization experiments
In the following section we show examples of the evolution of the selected molecules for the constrained optimization experiments. Figures 13, 14, and 15 show starting and final molecules, together with all molecules generated along the optimization path, and their values of penalized logP.
Unconstrained molecule optimization
Our architecture is tailormade for the scenario of constrained molecule optimization. However, as an additional task, we check what happens when we iteratively use the generator on the molecules being optimized. This should lead to diminishing similarity between the starting molecules and those in consecutive iterations. For the present task the set X needs to be a sample from the entire ZINC250K, whereas the set Y is chosen as a sample from the top20\(\%\) of molecules with the highest value of penalized logP. Each molecule is fed into the generator and the corresponding ‘optimized’ molecule’s latent space representation is obtained. The generated latent space representation is then treated as the new input for the generator. The process is repeated K times and the resulting set of molecules is \( \{ G(x),G(G(x))\}, \ldots \). Here, as in the previous task and as in previous research [30] we start the procedure from the 800 molecules with the lowest values of penalized logP in ZINC250K.
The results of our unconstrained molecule optimization are shown in Fig. 16. In Fig. 16a, c we observe that consecutive iterations keep shifting the distribution of the objective (penalized logP) towards higher values. However, the improvement from further iterations is decreasing. Interestingly, the maximum of the distribution keeps increasing (although in somewhat random fashion). After 10–20 iterations it reaches very high values of logP observed from molecules which are not druglike, similarly to those obtained with RL [35]. Both in the case of the RL approach and in our case, the molecules with the highest penalized logP after many iterations also become nondruglike—see Fig. 19 for a list of compounds with the maximum values of penalized logP in the iterative optimization procedure. This lack of druglikeness is related to the fact that after performing many iterations, the distribution of coordinates of our set of molecules in the latent space goes far away from the prior distribution (multivariate normal) used when training the JTVAE on ZINC250K. In Fig. 16b we show the evolution of the distribution of Tanimoto similarities between the starting molecules and those obtained after \(K = 1, 2, 5, 10\) iterations. We also show the similarity between the starting molecules and random molecules from ZINC250K. We observe that after 10 iterations the similarity between the starting molecules and the optimized ones is comparable to the similarity of random molecules from ZINC250K. After around 20 iterations the optimized molecules become less similar to the starting ones than random molecules from ZINC250K, as the set of optimized molecules is moving further away from the space of druglike molecules.
Molecular paths from unconstrained optimization experiments
In the following section we show examples of the evolution of selected molecules for the unconstrained optimization experiments. Figures 17 and 18 show starting and final molecules, together with all molecules generated during the iteration over the optimization path and their penalized logP values.
Molecules with the highest values of penalized logP
On Fig. 16c we plot the maximum value of penalized logP in the set of molecules being optimized as a function of number of iterations for unconstrained molecule optimization. In Fig. 19 we show corresponding molecules for iterations 1–24.
Activity
Lastly, we test compound activity optimization for the dopamine receptor D2, i.e. we want to increase the binding affinity of a compound towards DRD2. For this task we selected a set X of inactive compounds, and a set Y of active molecules which were extracted from the ChEMBL database. We used threshold of \({\rm K}_i < 100~{\rm nM}\) for selecting active compounds (2738 active compounds and 2254 inactive compounds were selected for training after filtering out duplicates).
For scoring the generated molecules, we trained a DRD2 activity prediction classification model based on ECFP fingerprints (generated with RDKit [50]). We chose to use a random forest model with 0.92 ROC AUC test score in threefold crossvalidation. In this task we also add 10 intermediate molecules from the optimization path to find more similar compound with improved activity. Table 7 quantitatively summarizes the experiment of activity optimization. Table 8 shows that the MolCycleGAN is able to increase activity of a selected inactive drug by a significant margin, based on the prediction of a bioactivity model. Figure 20 shows similarity of the optimized compounds to the starting molecules and compares their predicted activities. Examples of optimized compounds are presented in Fig. 21. To validate the results of the experiment, we performed docking procedures for a number of generated compounds and found that, on average, the optimized compounds have better docking energies than their progenitors (Fig. 22).
Conclusions
In this work, we introduce MolCycleGAN—a new model based on CycleGAN which can be used for the de novo generation of molecules. The advantage of the proposed model is the ability to learn transformation rules from the sets of compounds with desired and undesired values of the considered property. The model operates in the latent space trained by another model—in our work we use the latent space of JTVAE. The model can generate molecules with desired properties, as shown on the example of structural and physicochemical properties. The generated molecules are close to the starting ones and the degree of similarity can be controlled via a hyperparameter. In the task of constrained optimization of druglike molecules our model significantly outperforms previous results. In the future work we plan to extend the approach to multiparameter optimization of molecules using StarGAN [41]. It would also be interesting to test the model on cases where a small structural change leads to a drastic change in the property (e.g. the socalled activity cliffs) which are hard to model.
Availability of data and materials
All source code and datasets used to produce the reported results can be found online at: https://github.com/ardigen/molcyclegan.
Abbreviations
 CADD:

computeraided drug design
 VAE:

variational autoencoder
 GAN:

Generative Adversarial Networks
 RL:

Reinforcement Learning
 JTVAE:

Junction Tree Variational Autoencoder
 GCPN:

Graph Convolutional Policy Network
References
 1.
Ratti E, Trist D (2001) The continuing evolution of the drug discovery process in the pharmaceutical industry. Farmaco 56(1–2):13–19. https://doi.org/10.1016/S0014827X(01)010199
 2.
Rao VS, Srinivas K (2011) Modern drug discovery process: an in silico approach. J Bioinform Seq Anal 2(5):89–94
 3.
Bajorath J (2002) Integration of virtual and highthroughput screening. Nat Rev Drug Discov 1(11):882–894. https://doi.org/10.1038/nrd941
 4.
Lavecchia A, Di Giovanni C (2013) Virtual screening strategies in drug discovery: a critical review. Curr Med Chem 20(23):2839–2860
 5.
Honório KM, Moda TL, Andricopulo AD (2013) Pharmacokinetic properties and in silico adme modeling in drug discovery. J Med Chem 9(2):163–176
 6.
de Ruyck J, Brysbaert G, Blossey R, Lensink MF (2016) Molecular docking as a popular tool in drug design, an in silico travel. Adv Appl Bioinform 9:1–11
 7.
Segler MH, Preuss M, Waller MP (2018) Planning chemical syntheses with deep neural networks and symbolic ai. Nature 555(7698):604
 8.
Chen H, Engkvist O, Wang Y, Olivecrona M, Blaschke T (2018) The rise of deep learning in drug discovery. Drug Discov Today. https://doi.org/10.1016/j.drudis.2018.01.039
 9.
Duvenaud DK, Maclaurin D, Iparraguirre J, Bombarell R, Hirzel T, AspuruGuzik A, Adams RP (2015) Convolutional networks on graphs for learning molecular fingerprints. In: Advand neurology, pp. 2224–2232
 10.
Jastrzębski S, Leśniak D, Czarnecki WM (2016) Learning to smile (s). arXiv preprint arXiv:1602.06289
 11.
Coley CW, Barzilay R, Green WH, Jaakkola TS, Jensen KF (2017) Convolutional embedding of attributed molecular graphs for physical property prediction. J Chem Inf Model 57(8):1757–1772
 12.
Pham T, Tran T, Venkatesh S (2018) Graph memory networks for molecular activity prediction. arXiv preprint arXiv:1801.02622
 13.
Segler MH, Kogej T, Tyrchan C, Waller MP (2017) Generating focused molecule libraries for drug discovery with recurrent neural networks. ACS Cent Sci 4(1):120–131. https://doi.org/10.1021/acscentsci.7b00512
 14.
Bjerrum EJ, Threlfall R (2017) Molecular generation with recurrent neural networks (rnns). arXiv preprint arXiv:1705.04612
 15.
Winter R, Montanari F, Noé F, Clevert DA (2019) Learning continuous and datadriven molecular descriptors by translating equivalent chemical representations. Chem Sci. https://doi.org/10.1039/C8SC04175J
 16.
Gupta A, Müller AT, Huisman BJ, Fuchs JA, Schneider P, Schneider G (2018) Generative recurrent networks for de novo drug design. Mol Inform 37(1–2):1700111. https://doi.org/10.1002/minf.201700111
 17.
ArúsPous J, Blaschke T, Ulander S, Reymond JL, Chen H, Engkvist O (2019) Exploring the gdb13 chemical space using deep generative models. J Cheminform 11(1):20
 18.
Popova M, Isayev O, Tropsha A (2018) Deep reinforcement learning for de novo drug design. Sci Adv 4(7):7885
 19.
Kusner MJ, Paige B, HernándezLobato JM (2017) Grammar variational autoencoder. In: Proceedings of the 34th international conference on machine learning, volume 70, pp. 1945–1954
 20.
Dai H, Tian Y, Dai B, Skiena S, Song L (2018) Syntaxdirected variational autoencoder for structured data. arXiv preprint arXiv:1802.08786
 21.
ArúsPous J, Johansson S, Prykhodko O, Bjerrum EJ, Tyrchan C, Reymond JL, Chen H, Engkvist O (2019) Randomized SMILES strings improve the quality of molecular generative models. ChemRxiv
 22.
Olivecrona M, Blaschke T, Engkvist O, Chen H (2017) Molecular denovo design through deep reinforcement learning. J Cheminform 9(1):48
 23.
Li Y, Vinyals O, Dyer C, Pascanu R, Battaglia P (2018) Learning deep generative models of graphs. arXiv preprint arXiv:1803.03324
 24.
Li Y, Zhang L, Liu Z (2018) Multiobjective de novo drug design with conditional graph generative model. J Cheminform 10(1):33
 25.
Lim J, Hwang SY, Kim S, Moon S, Kim WY (2019) Scaffoldbased molecular design using graph generative model. arXiv preprint arXiv:1905.13639
 26.
Kingma DP, Welling M (2013) Autoencoding variational bayes. arXiv preprint arXiv:1312.6114
 27.
GómezBombarelli R, Wei JN, Duvenaud D, HernándezLobato JM, SánchezLengeling B, Sheberla D, AguileraIparraguirre J, Hirzel TD, Adams RP, AspuruGuzik A (2018) Automatic chemical design using a datadriven continuous representation of molecules. ACS Cent Sci 4(2):268–276
 28.
Samanta B, Abir D, Jana G, Chattaraj PK, Ganguly N, Rodriguez MG (2019) Nevae: a deep generative model for molecular graphs. Proc AAAI Conf Artif Intell 33:1110–1117
 29.
Simonovsky M, Komodakis N (2018) Graphvae: towards generation of small graphs using variational autoencoders. arXiv preprint arXiv:1802.03480
 30.
Jin W, Barzilay R, Jaakkola T (2018) Junction tree variational autoencoder for molecular graph generation. In: Dy, J., Krause, A. (eds.) Proceedings of the 35th international conference on machine learning. Proceedings of machine learning research, vol. 80. PMLR, Stockholmsmässan, Stockholm Sweden, pp. 2323–2332
 31.
Goodfellow I, PougetAbadie J, Mirza M, Xu B, WardeFarley D, Ozair S, Courville A, Bengio Y (2014) Generative adversarial nets. In: Advance neurology, pp. 2672–2680
 32.
Guimaraes GL, SanchezLengeling B, Outeiral C, Farias PLC, AspuruGuzik A (2017) Objectivereinforced generative adversarial networks (organ) for sequence generation models. arXiv preprint arXiv:1705.10843
 33.
SanchezLengeling B, Outeiral C, Guimaraes GL, AspuruGuzik A (2017) Optimizing distributions over molecular space. In: An objectivereinforced generative adversarial network for inversedesign chemistry (organic)
 34.
De Cao N, Kipf T (2018) Molgan: an implicit generative model for small molecular graphs. arXiv preprint arXiv:1805.11973
 35.
You J, Liu B, Ying Z, Pande V, Leskovec J (2018) Graph convolutional policy network for goaldirected molecular graph generation. In: Advances in neural information processing systems, pp. 6410–6421
 36.
Prykhodko O, Johansson S, Kotsias PC, ArúsPous J, Bjerrum EJ, Engkvist O, Chen H (2019) A de novo molecular generation method using latent vector based generative adversarial network
 37.
Zhu JY, Park T, Isola P, Efros AA (2017) Unpaired imagetoimage translation using cycleconsistent adversarial networks. In: Proceedings of the IEEE international conference on computer vision, pp. 2223–2232
 38.
Maziarka Ł, Pocha A, Kaczmarczyk J, Rataj K, Warchoł M (2019) Molcyclegan—a generative model for molecular optimization. In: Tetko IV, Kůrková V, Karpov P, Theis F (eds) Artificial neural networks and machine learning—ICANN 2019: Workshop and Special Sessions. Springer, Cham, pp 810–816
 39.
Weininger D (1988) Smiles, a chemical language and information system. 1. Introduction to methodology and encoding rules. J Chem Inf Comp Sci 28(1):31–36
 40.
Mao X, Li Q, Xie H, Lau RY, Wang Z, Paul Smolley S (2017) Least squares generative adversarial networks. In: 2017 IEEE international conference on computer vision (ICCV), pp. 2794–2802 https://doi.org/10.1109/ICCV.2017.304
 41.
Choi Y, Choi M, Kim M, Ha JW, Kim S, Choo J (2017) Stargan: Unified generative adversarial networks for multidomain imagetoimage translation. arXiv:1711.09020
 42.
Perarnau G, van de Weijer J, Raducanu B, Álvarez JM (2016) Invertible conditional gans for image editing. arXiv preprint arXiv:1611.06355
 43.
Sterling T, Irwin JJ (2015) Zinc 15ligand discovery for everyone. J Chem Inf Model 55(11):2324–2337. https://doi.org/10.1021/acs.jcim.5b00559
 44.
Rogers D, Hahn M (2010) Extendedconnectivity fingerprints. J Chem Inf Model 50(5):742–754. https://doi.org/10.1021/ci100050t
 45.
Gaulton A, Hersey A, Nowotka M, Bento AP, Chambers J, Mendez D, Mutowo P, Atkinson F, Bellis LJ, CibriánUhalte E, Davies M, Dedman N, Karlsson A, Magariños MP, Overington JP, Papadatos G, Smit I, Leach AR (2016) The ChEMBL database in 2017. Nucleic Acids Res 45(D1):945–954. https://doi.org/10.1093/nar/gkw1074; http://oup.prod.sis.lan/nar/articlepdf/45/D1/D945/8846762/gkw1074.pdf
 46.
Kingma DP, Ba J (2014) Adam: a method for stochastic optimization. arXiv preprint arXiv:1412.6980
 47.
Ioffe S, Szegedy C (2015) Batch normalization: accelerating deep network training by reducing internal covariate shift. In: Proceedings of the 32nd international conference on international conference on machine learning, volume 37. ICML’15, pp. 448–456. http://dl.acm.org/citation.cfm?id=3045118.3045167
 48.
Besnard J, Ruda GF, Setola V, Abecassis K, Rodriguiz RM, Huang XP, Norval S, Sassano MF, Shin AI, Webster LA (2012) Automated design of ligands to polypharmacological profiles. Nature 492(7428):215. https://doi.org/10.1038/nature11691
 49.
Bickerton GR, Paolini GV, Besnard J, Muresan S, Hopkins AL (2012) Quantifying the chemical beauty of drugs. Nat Chem 4(2):90
 50.
Landrum G (2016) Rdkit: Opensource cheminformatics software
Acknowledgements
We would like to thank Sabina Podlewska for her helpful comments and for fruitful discussions.
Funding
Not applicable.
Author information
Affiliations
Contributions
LM and AP derived the concept. LM wrote relevant code and performed the experimental work and analysis. AP and JK wrote the paper. KR provided feedback and critical input. TD prepared the revised version of the manuscript. MW did a critical revision of the manuscript. All authors read and approved the final manuscript.
Corresponding author
Correspondence to Łukasz Maziarka.
Ethics declarations
Competing interests
The authors declare that they have no competing interests.
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated in a credit line to the data.
About this article
Cite this article
Maziarka, Ł., Pocha, A., Kaczmarczyk, J. et al. MolCycleGAN: a generative model for molecular optimization. J Cheminform 12, 2 (2020). https://doi.org/10.1186/s1332101904041
Received:
Accepted:
Published:
Keywords
 Drug design
 Molecular optimization
 Generative models
 Deep learning