The olfactory system employs an ensemble of odorant receptors (ORs) to sense molecules and to derive olfactory percepts. We hypothesized that ORs may be considered 3D spatial filters that extract molecular features relevant to the olfactory system, similar to the spatial filters employed in other modalities. If so, the composition of OR ensemble can be understood by training such filters using conventional artificial intelligence methods and large-scale databases of 3D molecular structures. We trained artificial neural networks to represent the chemical space of odorants and used this representation to predict human olfactory percepts. First, we trained an autoencoder called DeepNose to deduce a compressed representation of odorant molecules based on their 3D spatial structure. Next, we tested the ability of DeepNose features in predicting human odorant percepts based on 3D molecular structures alone. Finally, we finetuned the DeepNose network to better represent perceptual properties of odorants defined as semantic descriptors. We found that, despite the lack of human expertise, DeepNose features led to perceptual predictions of comparable or higher accuracy to molecular descriptors often used in computational chemistry. We propose that DeepNose network can use 3D molecular structures to yield high-quality predictions of human olfactory percepts and can help understand the factors influencing the composition of ORs ensemble.