Oligonucleotides are commonly purified using either ion exchange chromatography (IEX) or ion-pair reversedphase liquid chromatography (IP-RPLC). This study compares the purification of a crude 20-mer oligonucleotide (ON) using both methods under preparative conditions. Two variables were investigated during the separation: column load and gradient slope. Although the IEX purifications using agarose-based resins had longer cycle times, this was compensated by the high loadability compared to the silica-based IP-RPLC media. This resulted in both higher productivity and lower solvent consumption at all evaluated purities, ranging from 95 % to 99 %, at optimal productivity levels. At 95 % purity, IEX achieved more than twice the productivity, and at 99 % purity, the productivity was seven times higher. Additionally, solvent consumption was significantly reduced, with IEX consuming only one-third to one-tenth of the solvents compared to IP-RPLC at purities from 95 % to 99 %.

Poor data quality limits the advantageous power of Machine Learning (ML) and weakens high-performing ML software systems. Nowadays, data are more prone to the risk of poor quality due to their increasing volume and complexity. Therefore, tedious and time-consuming work goes into data preparation and improvement before moving further in the ML pipeline. To address this challenge, we propose an intelligent data-centric evaluation framework that can identify high-quality data and improve the performance of an ML system. The proposed framework combines the curation of quality measurements and unsupervised learning to distinguish high- and low-quality data. The framework is designed to integrate flexible and general-purpose methods so that it is deployed in various domains and applications. To validate the outcomes of the designed framework, we implemented it in a real-world use case from the field of analytical chemistry, where it is tested on three datasets of anti-sense oligonucleotides. A domain expert is consulted to identify the relevant quality measurements and evaluate the outcomes of the framework. The results show that the quality-centric data evaluation framework identifies the characteristics of high-quality data that guide the conduct of efficient laboratory experiments and consequently improve the performance of the ML system. 

This study presents an improved workflow to support the development of machine learning models to predict oligonucleotide retention times, peak widths and thus peak resolutions, from larger datasets where manual processing is not feasible. We explored diverse oligonucleotide forms, ranging from native to fully phosphorothioated, using three different gradient slopes. Both native and phosphorothioated oligonucleotides were separated, using a chromatographic C18 system with tributylaminium ion as the ion-pair reagent in the eluent, resulting in retention time data for approximately 900 sequences per gradient. For managing the large and extensive datasets, we developed a semi-automatic rule-based approach for retention time determination, peak decomposition, peak width assessment, signal-to-noise ratio, and skewness analysis. Probability density functions (PDFs) were fitted to elution profiles, with PDF selection based on an Ftest. Co-eluting peaks were addressed using a multiple Gaussian PDF. The encoded sequence data underwent modeling using support vector regression (SVR), gradient boosting (GB), random forest (RF), and decision tree (DT) models. GB and SVR showed promise for retention predictions, while RT and DT were faster but demonstrated limited generalization capabilities. The machine learning models exhibited larger errors for the shallowest gradient and lower predictability for P=O sequences, potentially due to signal intensity and sequence heterogeneity. Improvements in signal-to-noise ratios were considered, including mass spectrometry in selected ion monitoring mode. The best model for this data sets were GB, closely followed by the SVR model. With established models for retention and peak width, chromatograms can now be predicted for various gradient slopes, offering prediction of impurity peak resolution for arbitrary sequences and gradient slopes.

This work introduces the Adsorption Energy Distribution (AED) calculation using competitive adsorption isotherm data, enabling investigation of the simultaneous AED of two components for the first time. The AED provides crucial insights by visualizing competitive adsorption processes, offering an alternative adsorption isotherm model without prior assuming adsorption heterogeneity, and assisting in model selection for more accurate retention mechanistic modeling. The competitive AED enhances our understanding of multicomponent interactions on stationary phases, which is crucial for understanding how analytes compete on the stationary phase surface and for selecting adsorption models for numerical optimization of preparative chromatography. Here, the two-component AED was tested on both synthetic and experimental data, and a very successful outcome was achieved.

In indirect detection, sample components lacking detectable properties are detected by adding a detectable component to the eluent, a so-called probe that interacts with the analytes to be detected. This study focuses on modeling indirect detection in two principally different cases. In case (1), the analyte component has the same charge as the probe component, so the probe acts as a co-ion of the analyte. In case (2), the analyte component has the opposite charge to the probe, so the probe acts as a counter-ion of the analyte. In the co-ion case (1), the analytes are alkyl sulfonates, and a competitive bi-Langmuir isotherm model was used. In the counter-ion case (2), the analytes are amines, and a modified bi-Langmuir isotherm model, incorporating ion-pairing on the stationary phase surface, was derived and applied for simulating the elution profiles. The chromatographic system comprised an XBridge Phenyl column as the stationary phase and an acetonitrile/phosphate buffer mixture with varying concentrations of sodium 2-naphthalenesulfonate as the eluent. In both cases, the detectable probe component was sodium 2-naphthalenesulfonate. The applied isotherm models successfully predicted system peaks with high agreement in both model cases, with calculated relative errors in retention times typically below 4.72 % and often below 1 %. The models were employed to predict the sensitivity of analytical methods, demonstrating excellent agreement between experimental and calculated sensitivities. These findings confirm the validity of the new adsorption isotherm model under these experimental conditions. 

Short interfering RNA (siRNA) represents a rapidly expanding class of marketed oligonucleotide therapeutics. Due to its double-stranded nature, the characterization of siRNA is twofold: (i) at the single-strand (denaturing) level for impurity profiling and (ii) at the intact (nondenaturing) level to confirm duplex formation and quantify excess single strands (including single strand-derived impurities). While denaturing analysis can be carried out using conventional ion-pair reversed-phase liquid chromatography (IP-RPLC), nondenaturing characterization of siRNA is a significantly less straightforward task. Typical IP-RPLC conditions have an intrinsic denaturing effect on siRNA, thereby limiting the development of viable approaches for the intact duplex analysis. In this study, we demonstrate, through the design of experiments of siRNA melting temperatures and chromatography analyses, that the simple addition of salts, such as phosphate-buffered saline and ammonium acetate, to eluents enhances the suitability of IP-RPLC for the nondenaturing analysis of siRNA during both UV- and mass spectrometry-based analysis. This work represents a milestone in overcoming the challenges associated with nondenaturing analysis of siRNAs by IP-RPLC and offers a fresh angle for exploring IP-RPLC of siRNAs.

The retention of three peptides was studied under analytical and overloaded conditions at different concentrations of trifluoroacetic acid (TFA) and water added to the co-solvent methanol (MeOH). Four columns with different stationary phase properties, i.e., silica, diol, 2-ethylpyridine and cyanopropyl (CN) columns, were evaluated in this investigation. The overall aim was to get a deeper understanding on how column chemistry as well as water and TFA in the co-solvent affect the analytical and overloaded elution profiles using multivariate design of experiments and adsorption measurements of co-solvent components. Multivariate experimental design modeling indicated that water had on average around five times higher effect on the retention than the addition of TFA. The results also showed that the retention increases with the addition of TFA and water to the co-solvent on all columns except the CN column, on which the retention decreased. When examining the effect of adding water to the co-solvent, evidence of a hydrophilic interaction liquid chromatography (HILIC)-like retention mechanism was found on the three other columns with more polar stationary phases. However, on the CN column water acted as an additive, decreasing the retention due to competition with the peptide for available adsorption surface. Adsorption isotherm measurements of the polar solvent MeOH showed that MeOH adsorbs much weaker on the CN column than on the other columns. Addition of TFA and water to the co-solvent substantially sharpened the elution profiles under both overloaded and analytical conditions. Adding a small amount of TFA (from 0 % to 0.05 %) to the co-solvent substantially improved the peak shape of the elution profiles, while further addition (from 0.05 % to 0.15 %) had only a minor effect on the elution profile shape. The reduced retention on the CN column could not be explained by TFA adsorption, which was very weak on all studied columns (retention factor, 0.05–0.15). One could therefore speculate that the ion-pairing complex formed between the peptide and TFA in the mobile phase, reduce the retention due to its reduced polarity. On the other columns displaying HILIC-like properties, the TFA probably just decreased the pH of the mobile phase, thereby promoting the partitioning of the peptide into the water-rich layer. Finally, peak deformation due to diluent–eluent mismatch was observed under overloaded conditions. This was most severe in the cases where MeOH adsorption to the stationary phase was strong and the peptides were only mildly retained. Adding 1,4-dioxan to the diluent resolved this issue. 

Creating resilient machine learning (ML) systems has become necessary to ensure production-ready ML systems that acquire user confidence seamlessly. The quality of the input data and the model highly influence the successful end-to-end testing in data-sensitive systems. However, the testing approaches of input data are not as systematic and are few compared to model testing. To address this gap, this paper presents the Fault Injection for Undesirable Learning in input Data (FIUL-Data) testing framework that tests the resilience of ML models to multiple intentionally-triggered data faults. Data mutators explore vulnerabilities of ML systems against the effects of different fault injections. The proposed framework is designed based on three main ideas: The mutators are not random; one data mutator is applied at an instance of time, and the selected ML models are optimized beforehand. This paper evaluates the FIUL-Data framework using data from analytical chemistry, comprising retention time measurements of anti-sense oligonucleotide. Empirical evaluation is carried out in a two-step process in which the responses of selected ML models to data mutation are analyzed individually and then compared with each other. The results show that the FIUL-Data framework allows the evaluation of the resilience of ML models. In most experiments cases, ML models show higher resilience at larger training datasets, where gradient boost performed better than support vector regression in smaller training sets. Overall, the mean squared error metric is useful in evaluating the resilience of models due to its higher sensitivity to data mutation. 

This study investigated the influence of pH on the retention of solutes using a mixed-mode column with carboxyl (-COOH) groups acting as weak cation exchanger bonded to the terminal of C18 ligands (C18-WCX column) and a traditional reversed-phase C18 column. First, a model based on electrostatic theory was derived and successfully used to predict the retention of charged solutes (charged, and ionizable) as a function of mobile phase pH on a C18-WCX column. While the Horváth model predicts the pH-dependent retention of ionizable solutes in reversed-phase liquid chromatography (RPLC) solely based on solute ionization, the developed model incorporates the concept of surface potential generated on the surface of the stationary phase and its variation with pH. To comprehensively understand the adsorption process, adsorption isotherms for these solutes were individually acquired on the C18-WCX and reversed-phase C18 columns. The adsorption isotherms followed the Langmuir model for the uncharged solute and the electrostatically modified Langmuir model for charged solutes. The elution profiles for the single components were calculated from these isotherms using the equilibrium dispersion column model and were found to be in close agreement with the experimental elution profiles. To enable modelling of two-component cases involving charged solute(s), a competitive adsorption isotherm model based on electrostatic theory was derived. This model was later successfully used to calculate the elution profiles of two components for scenarios involving (a) a C18 Column: two charged solutes, (b) a C18 Column: one charged and one uncharged solute, and (c) a C18-WCX Column: two charged solutes. The strong alignment between the experimental and calculated elution profiles in all three scenarios validated the developed competitive adsorption model. 

In this study, overloaded elution profiles under ultra-high-pressure liquid chromatographic (UHPLC) conditions and accounting for the severe pressure and temperature gradients generated, are compared with experimental data. The model system consisted of an C18 column packed with 1.7-µm particles (i.e., a UHPLC column) and the solute was 1,3,5-tri‑tert-butylbenzene eluted with a mobile phase composed of 85/15 (v/v) acetonitrile/water. Two thermal modes were considered, and the solute was eluted at the very high inlet pressures necessary to achieve a highly efficient and rapid chromatographic process, as provided by using columns packed with small particles. However, the high inlet pressure and high linear velocity of the mobile phase caused the production of a significant amount of heat, and consequently, the formation of axial and radial temperature gradients. Due to these gradients, the retention and the mobile phase velocity were no longer constant. Thus, simple mathematical models consisting only of the mass balance equations are unsuitable to properly model the elution profiles. Here, the elution concentration profiles were predicted using a combined two-dimensional heat and mass transfer model, also including the calculation of the mobile phase velocity distribution. The isotherm adsorption model was the bi-Langmuir isotherm model with Henry constants that depended on the local temperature and pressure in the column. These adjustments allowed us to precisely account for changes in the shape and retention of the overloaded concentration profiles in the mobile phase. The proposed model provided accurate predictions of the overloaded concentration profiles, demonstrating good agreement with experimental profiles eluted under severe pressure and temperature gradients in the column even in the most extreme cases where the pressure drops reached 846 bar and the temperature gradients equaled 0.15 K mm−1 and 0.95 K mm−1 in the axial and the radial directions, respectively. In such cases 36 % decrease of the retention factor was observed along the column and 2 % increase in radial direction. These changes, combined with the velocity distribution, shifted the overloaded elution profile’s shock towards the center of the column, advancing approximately 3 mm from its initial position close to the column wall. Ultimately, this resulted in the broadening of the elution band.