A greedy stacking algorithm for model ensembling and domain weighting.

OBJECTIVE: Because it is impossible to know which statistical learning algorithm performs best on a prediction task, it is common to use stacking methods to ensemble individual learners into a more powerful single learner. Stacking algorithms are usually based on linear models, which may run into problems, especially when predictions are highly correlated. In this study, we develop a greedy algorithm for model stacking that overcomes this issue while still being very fast and easy to interpret. We evaluate our greedy algorithm on 7 different data sets from various biomedical disciplines and compare it to linear stacking, genetic algorithm stacking and a brute force approach in different prediction settings. We further apply this algorithm on a task to optimize the weighting of the single domains (e.g., income, education) that build the German Index of Multiple Deprivation (GIMD) to be highly correlated with mortality. RESULTS: The greedy stacking algorithm provides good ensemble weights and outperforms the linear stacker in many tasks. Still, the brute force approach is slightly superior, but is computationally expensive. The greedy weighting algorithm has a variety of possible applications and is fast and efficient. A python implementation is provided.

2,8-Disubstituted-1,6-Naphthyridines and 4,6-Disubstituted-Isoquinolines with Potent, Selective Affinity for CDK8/19.

We demonstrate a designed scaffold-hop approach to the discovery of 2,8-disubstituted-1,6-naphthyridine- and 4,6-disubstituted-isoquinoline-based dual CDK8/19 ligands. Optimized compounds in both series exhibited rapid aldehyde oxidase-mediated metabolism, which could be abrogated by introduction of an amino substituent at C5 of the 1,6-naphthyridine scaffold or at C1 of the isoquinoline scaffold. Compounds 51 and 59 were progressed to in vivo pharmacokinetic studies, and 51 also demonstrated sustained inhibition of STAT1(SER727) phosphorylation, a biomarker of CDK8 inhibition, in an SW620 colorectal carcinoma human tumor xenograft model following oral dosing.

Discovery of Potent, Selective, and Orally Bioavailable Small-Molecule Modulators of the Mediator Complex-Associated Kinases CDK8 and CDK19.

The Mediator complex-associated cyclin-dependent kinase CDK8 has been implicated in human disease, particularly in colorectal cancer where it has been reported as a putative oncogene. Here we report the discovery of 109 (CCT251921), a potent, selective, and orally bioavailable inhibitor of CDK8 with equipotent affinity for CDK19. We describe a structure-based design approach leading to the discovery of a 3,4,5-trisubstituted-2-aminopyridine series and present the application of physicochemical property analyses to successfully reduce in vivo metabolic clearance, minimize transporter-mediated biliary elimination while maintaining acceptable aqueous solubility. Compound 109 affords the optimal compromise of in vitro biochemical, pharmacokinetic, and physicochemical properties and is suitable for progression to animal models of cancer.

An approach to the core structure of leiodermatolide.

The synthesis of the 16-membered core structure of leiodermatolide 40 has been achieved in 26 linear steps starting from (R)-Roche ester. The key steps in the synthesis of 40 are a Stille cross-coupling between two main fragments 11 and 33 having roughly equal size. For the trisubstituted C4/C5 double bond a carbometalation reaction followed by a Suzuki coupling was used. A Yamaguchi macrolactonization furnished macrolactone 39.

Probing the influence of an allylic methyl group in zearalenone analogues on binding to Hsp90.

By the replacement of an acetate with propionate by means of organic synthesis, a range of zearalenone analogues were prepared that feature an allylic methyl group. For the synthesis of the aliphatic region of the analogues, we used an asymmetric alkylation to yield pentenol derivatives 16 and ent-16. By means of hydroboration the corresponding aldehydes were secured. These were coupled with 2-pentynol derivate 23 by means of a Carreira acetylide addition. Further routine steps led to the sulfones 29 and 45, respectively. After merging them with 2-bromobenzaldehyde 9 in a Julia-Kocienski reaction, metalation, carboxylation, and protecting-group manipulations gave the seco acids 35 and 49. By means of lactonization under Mitsunobu (alcohol activation) or Trost-Kita conditions (carboxyl activation), all four possible macrocyclic ketone stereoisomers were accessible. In all, considering various protecting-group decorations, 16 analogues were obtained and tested for cytotoxicity (L929 mouse fibroblast cell line). Whereas most of the analogues were less active than zearalenone (IC(50)=9.4 µM), the resorcinol derivatives were comparable, with one stereoisomer (40 b) being slightly more active (IC(50)=6.6 µM). These results were also reflected in the binding assays to Hsp90 in which 40 b showed a dissociation constant (K(d)) value of 130 nM.