Poisson Statistics of Combinatorial Library Sampling Predict False Discovery Rates of Screening

Abstract

Microfluidic droplet-based screening of DNA-encoded one-bead-one-compound combinatorial libraries is a miniaturized, potentially widely distributable approach to small molecule discovery. In these screens, a microfluidic circuit distributes library beads into droplets of activity assay reagent, photochemically cleaves the compound from the bead, then incubates and sorts the droplets based on assay result for subsequent DNA sequencing-based hit compound structure elucidation. Pilot experimental studies revealed that Poisson statistics describe nearly all aspects of such screens, prompting the development of simulations to understand system behavior. Monte Carlo screening simulation data showed that increasing mean library sampling (ε), mean droplet occupancy, or library hit rate all increase the false discovery rate (FDR). Compounds identified as hits on k > 1 beads (the replicate k class) were much more likely to be authentic hits than singletons (k = 1), in agreement with previous findings. Here, we explain this observation by deriving an equation for authenticity, which reduces to the product of a library sampling bias term (exponential in k) and a sampling saturation term (exponential in ε) setting a threshold that the k-dependent bias must overcome. The equation thus quantitatively describes why each hit structure’s FDR is based on its k class, and further predicts the feasibility of intentionally populating droplets with multiple library beads, assaying the micromixtures for function, and identifying the active members by statistical deconvolution.

Randomness in DNA Encoded Library Selection Data Can Be Modeled for More Reliable Enrichment Calculation

“…A recent report regarding bead-displayed DELs in a microfluidic screening format reported a similar assessment, in which four beads observed together in a droplet set a baseline for a satisfactory false-discovery rate. 21 By comparing multiple replicated DEL selections, we confirmed that the random noise of DEL data universally and consistently follows the Poisson distribution. The only ways to reduce the Poisson noise are to conduct a large number of repeated selections or to increase the sampling ratio.…”

“…A recent report regarding bead-displayed DELs in a microfluidic screening format reported a similar assessment, in which four beads observed together in a droplet set a baseline for a satisfactory false-discovery rate. 21…”

“…20 MacConnell and Paegel demonstrated that the false discovery rate arising from bead-displayed DELs in a microfluidic screening format is associated with the number of beads observed together in a droplet, a situation that is mathematically analogous to the copy counts in other DEL observations. 21 There is currently no consensus on how to handle low copy counts. Many choose to either directly compare copy numbers or linearly transform copy numbers before applying cutoffs, which can often be arbitrary and inconsistent.…”

DNA-encoded library versus RNA-encoded library selection enables design of an oncogenic noncoding RNA inhibitor

“…Our approach integrates two orthogonal and yet highly complementary library analyses, deploying evolutionary principles (56)(57)(58)(59) at the earliest stage of probe discovery. Screening hits are identified with confidence by pairing redundant hit isolation in FACS (27) with in vitro selection-based RNA motif ranking by sequence abundance (20).…”

“…1D; SI Appendix, Fig. S5) (22,27). Each library member was screened as a mixture of two Pro(N 3 ) diastereomers, thus each hit was synthesized as optically pure compounds in diastereomer pairs (e.g., compounds 1 and 2).…”