Deep learning models facilitate the discovery of molecules with tailored properties among billions of candidate compounds. However, the computational burden to develop and deploy state-of-the-art models continuously increases, limiting their scalability. Most large-scale models are unimodal in nature and overlook the potential to leverage complementary molecular data modalities. To address these shortcomings, this paper introduces the Graph-Language Alignment for Chemical Inference and Exploration using Representations (GLACIER) model, a student-teacher framework that integrates molecular graphs, SMILES strings, and physicochemical descriptors to learn rich molecular embeddings. Our framework consists of three stages: (1) we pretrain three student encoders on 100,000 drug-like molecules: a message-passing neural network for molecular graphs, a transformer-based encoder for SMILES strings, and a multilayer perceptron for physicochemical descriptors, (2) we fuse these student modalities using a novel Finsler geometry-aware module, and (3) distill complementary knowledge from large teacher models, including MiniMol and MolFormer, into a single lightweight model via contrastive learning. We demonstrate that GLACIER is a robust framework that delivers high predictive performance and computational efficiency in complex molecular property prediction tasks. Our code is publicly available at this https URL.

The RNA world hypothesis suggests a pathway of how life emerged on early earth. It assumes that life started with RNA based systems, capable of storing, transmitting and replicating information, envisioning that monomers and short RNA oligomers interact to form longer strands, eventually becoming catalytically active ribozymes. Key reactions in RNA pools are hybridization, dehybridization, templated ligation, and cleavage. Those reactions depend on many environmental parameters and the wide range of possible configurations among interacting strands. In order to scan such high dimensional parameter spaces, efficient descriptions are needed. Motif rate equations project complex strand reactor dynamics onto sequence motif space. Here we present a Bayesian inference framework to infer their parameters from ligation count data produced by strand reactor simulations. This provides a framework to match the simpler motif rate equations to more complex simulations. Additionally, it is a step towards inferring reaction rate constants directly from experimental data, including rigorous uncertainty estimation. This could be an essential procedure to connect theory and experiment, and deepen our understanding of the essential features necessary for life to emerge.