25 January 2016 in Journal of Cheminformatics
Hierarchical cluster analysis (HCA) is a widely used classificatory technique in many areas of scientific knowledge. Applications usually yield a dendrogram from an HCA run over a given data set, using a grouping algorithm and a similarity measure. However, even when such parameters are fixed, ties in proximity (i.e. two equidistant clusters from a third one) may produce several different dendrograms, having different possible clustering patterns (different classifications). This situation is usually disregarded and conclusions are based on a single result, leading to questions concerning the permanence of clusters in all the resulting dendrograms; this happens, for example, when using HCA for grouping molecular descriptors to select that less similar ones in QSAR studies. Representing dendrograms in graph theoretical terms allowed us to introduce four measures of cluster frequency in a canonical way, and use them to calculate cluster frequencies over the set of all possible dendrograms, taking all ties in proximity into account. A toy example of well separated clusters was used, as well as a set of 1666 molecular descriptors calculated for a group of molecules having hepatotoxic activity to show how our functions may be used for studying the effect of ties in HCA analysis. Such functions were not restricted to the tie case; the possibility of using them to derive cluster stability measurements on arbitrary sets of dendrograms having the same leaves is discussed, e.g. dendrograms from variations of HCA parameters. It was found that ties occurred frequently, some yielding tens of thousands of dendrograms, even for small data sets. Our approach was able to detect trends in clustering patterns by offering a simple way of measuring their frequency, which is often very low. This would imply, that inferences and models based on descriptor classifications (e.g. QSAR) are likely to be biased, thereby requiring an assessment of their reliability. Moreover, any classification of molecular descriptors is likely to be far from unique. Our results highlight the need for evaluating the effect of ties on clustering patterns before classification results can be used accurately.
22 July 2022 in Proceedings of the National Academy of Sciences
The periodic system, which intertwines order and similarity among chemical elements, arose from knowledge about substances constituting the chemical space. Little is known, however, about how the expansion of the space contributed to the emergence of the system—formulated in the 1860s. Here, we show by analyzing the space between 1800 and 1869 that after an unstable period culminating around 1826, chemical space led the system to converge to a backbone structure clearly recognizable in the 1840s. Hence, the system was already encoded in the space for about two and half decades before its formulation. Chemical events in 1826 and in the 1840s were driven by the discovery of new forms of combination standing the test of time. Emphasis of the space upon organic chemicals after 1830 prompted the recognition of relationships among elements participating in the organic turn and obscured some of the relationships among transition metals. To account for the role of nineteenth century atomic weights upon the system, we introduced an algorithm to adjust the space according to different sets of weights, which allowed for estimating the resulting periodic systems of chemists using one or the other weights. By analyzing these systems, from Dalton up to Mendeleev, Gmelin’s atomic weights of 1843 produce systems remarkably similar to that of 1869, a similarity that was reinforced by the atomic weights on the years to come. Although our approach is computational rather than historical, we hope it can complement other tools of the history of chemistry.
7 October 2022 in ChemRxiv
The Periodic Table (PT) is perhaps the most famous and widespread icon of chemistry. It orders chemical elements by their nuclear charge and groups them into families according to their similarity. Such arrangement was built using data about formulae of few compounds available in 19th century. Since then, the number of compounds has grown exponentially during the 20th and 21st centuries, and new types of compounds have been obtained that were unknown to pioneers, rising the question about the validity and generality of the PT. Can these patterns be extracted from current data or are they constrained to a particular chemical domain? To answer this question we conducted a Big Data exploration of chemical similarity using formulae of compounds reported since around 1800. We found that the similarities between elements of the same family are resilient to attacks and are ubiquitous along chemical contexts. We also found that PT groups approach true equivalence classes, being the most symmetrical and transitive on our data. These features point to an underlying structure in the PT ruling the similarity between elements, which agrees with its fundamental nature. Time analysis revealed that since around 1980 all similarity relations are waning by an increasing production of unique formulae on almost all elements, leading to a singularization of elements. Nonetheless, PT families tend to be more frequently found, showing they prevail over any other similarity pattern.
18 September 2024 in Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences
The periodic table organizes chemical elements into families based on their similarity, now understood through Quantum Mechanics. However, these families were inferred from limited compounds in the nineteenth century. Since then, the number of compounds has exponentially grown, leading to the discovery of new types unknown to pioneers. This situation prompts the question of whether these families can still be discerned from the amassed data, or if their recognition is confined to specific thermochemical domains. To address this inquiry, we conducted a comprehensive exploration by comparing formulae (1771–2015) as a proxy for chemical similarity. Our findings reveal that stoichiometry not only captures a significant portion of the trends observed within families but also unveils other intriguing features of the formal structure of chemical similarity. These patterns approach equivalence classes independent of thermochemical context and demonstrate high resilience. Temporal analysis demonstrates that, since approximately 1980, similarity is diminishing due to an increasing production of unique formulae for nearly all elements. Nevertheless, chemical families endure over time and they stand out as the most robust similarity patterns. Our analysis offers compelling evidence that any study will reach the same conclusions, provided there is a sufficient diversity in input compound data.