TMIC Wishart Node’s Post

We are thrilled to share the launch of OligoDistiller, a MS-platform agnostic, multifunctional R package and web tool designed to tackle some of the most challenging data aspects of oligonucleotide analysis! 🌟🧬 In our latest article, we present how OligoDistiller facilitates the interpretation of complex LC-HRMS data for oligonucleotides—helping researchers better understand critical quality attributes like impurity profiles, stability, metabolites, and sequence conformity. Key features of OligoDistiller include:  ✔️Works for centroided MS1 and MS2 data in both negative and positive ion mode. ✔️ MS1: Annotation & quantification of 45 oligonucleotide-related LC-MS features, including 13 unknown impurities and 6 overlapping isotope envelopes (OIEs), covering 90.8% of detected peaks. ✔️ MS2: Major product ions assigned from MS2 spectra for a 47-mer DNA strand, providing 80.3% coverage of the oligonucleotide sequence. ✔️ An interactive, intuitive data review module, allowing users to inspect raw spectra linked to each feature. ✔️ Detailed isotope quality metrics for all annotated features. What makes OligoDistiller even more accessible is that it caters to both programming and non-programming users! Thanks to a detailed tutorial, anyone—regardless of technical background—can easily use the tool and analyze their data effectively. 🔗 Check out the tool and these tutorials here: @https://lnkd.in/eqKWMhvj Let’s push the boundaries of oligonucleotide analysis together! 💡 #oligonucleotide #MassSpectrometry #LCMS #OligoDistiller #Bioinformatics #DataAnalysis #Deconvolution #PharmaInnovation #ResearchTools #TutorialForAll Thomas De Vijlder Jennifer Lippens, Ph. D

𝐁𝐨𝐥𝐭𝐳-1 𝐦𝐚𝐭𝐜𝐡𝐞𝐬 𝐀𝐥𝐩𝐡𝐚𝐅𝐨𝐥𝐝3 𝐩𝐞𝐫𝐟𝐨𝐫𝐦𝐚𝐧𝐜𝐞 𝐰𝐢𝐭𝐡 𝐧𝐨 𝐮𝐬𝐚𝐠𝐞 𝐥𝐢𝐦𝐢𝐭𝐚𝐭𝐢𝐨𝐧𝐬. We were first to make it available, and now have an API and large throughput support for it as well! See our blogpost for more detail: https://lnkd.in/gCXuzmEh The authors find that Boltz-1 matches the performance of Chai-1, and therefore AlphaFold3 on protein, protein-ligand, and protein-dna/rna complexes. Lots of folks have been asking about AF3 for their use cases, so very interested to hear how well Boltz works on AbAg complexes, enzymes and protein-small molecule systems. Get in touch to learn more! Try it out: https://lnkd.in/gqnAG8Tz

Calibration Science, Part II: Systematic Error, Signal-to-Noise Ratios, and How to Reduce Random Error #cannabisanalysis #cannabisscience https://hubs.la/Q02vMkRF0

🌟 Addressing Data Biases in Machine Learning for Protein-Ligand Interaction Predictions: The Case of BindingNet 🌟 In the field of machine learning models for predicting protein-ligand interactions, much of the focus has been on enhancing model architectures and optimizing performance metrics on public datasets. However, real-world applications require extra attention to the underlying biases in the training data to prevent models from exhibiting unintended behaviors. 😵💫 One interesting case of addressing such biases comes from the BindingNet project. In their study, the researchers discovered that models trained on public datasets like PDBbind often displayed a strong positive correlation between the predicted binding affinities and solvent-accessible surface area (SASA). While SASA is a relevant feature for protein-ligand binding, this reliance can lead models to overly focus on surface area interactions at the expense of other important factors. As a result, models trained on such biased data may perform well in benchmarks but struggle in practical applications where the relationship between SASA and binding affinity is less pronounced. 👯 The BindingNet team tackled this issue by constructing a large dataset of 69,816 high-quality protein-ligand complexes, using comparative complex structure modeling. They expanded the structural diversity of ligands in the dataset, introducing compounds with varying SASA values. This strategy helped reduce the over-reliance on SASA, allowing the machine learning models to learn more generalizable binding interactions. By enriching the dataset with ligands that don’t follow a simple SASA-binding affinity correlation, the model became more robust and capable of capturing a wider range of molecular interactions. This is a significant step forward in improving the generalization capacity of models beyond typical benchmarks. 📄 This work exemplifies a data-first approach to tackling real-world challenges in model reliability. Instead of concentrating solely on refining model architecture, the researchers looked deeper into the training data, identifying and addressing biases. For those interested in how data science and drug discovery intersect, this research offers a valuable example of how addressing biases in the data itself can lead to machine learning models that perform more reliably in real-world applications. 👍 👉 Check out the work here: https://lnkd.in/gmrsyu7E #DrugDiscovery #DrugDesign #MachineLearning #AIforScience

The benefits of using BioloMICS for biological data management. Combining the ease of MS-Excel with the power of true databases, BioloMICS offers a comprehensive solution for managing and analyzing data efficiently. With various versions available, tailored to different business needs, BioloMICS provides a flexible and cost-effective option for handling biological data. Additionally, the user-friendly interface ensures a smooth experience for users. In the realm of bioinformatics, support plays a crucial role in addressing issues promptly and effectively. BioloMICS stands out by offering reliable assistance to users, ensuring a seamless experience in managing bioinformatics-related challenges. Check out our website for more information: www.bio-aware.com

Unleash the Power of #PPI3D: A Comprehensive Web Server to Explore, Analyze, and Model Protein-Protein, Protein-Peptide, and Protein-Nucleic Acid Interactions With interactive analysis tools and a regularly updated database, the updated PPI3D web server empowers scientists to gain deeper insights into biomolecular mechanisms. Quick Read: https://lnkd.in/drTHtmRw #bioinformatics #ppi #proteininteractions #proteinmodeling #structuralbiology #sciencenews #biotechnology

Sriya.AI ran a 2nd set of data on Percent Protein Match based on Genome sequence and the results are as under: We used our Target Variable as: %match: Binary classification based on the threshold: Match: %match > 99.6% Not match: %match <= 99.6%. We are using Not Matched as the outcome to be improved or here needs reduction Sriya.AI uses a unique 4 step process with its proprietary SXI++ Algorithm suite Score: We convert all independent variables into 1 “Super Feature = SXI++”. This makes a complex multi variate problem into a simpler bi variate solution Total Features (Post-Feature Encoding): 184 SXI++ = 2.24 SXI > 2.24 resulted in 100% Match of %Match > 99.6 data. This establishes SXI as a perfect delineator of Match vs Not Match. Correlate: We then correlate SXI++ to Target Feature/variable The relationship was nonlinear (2nd order) The correlation between SXI and Not Match is 0.99 Correlation was negative or SXI++ needs to go up to reduce Not Match Predict: What was the prediction accuracy, precision, recall and AUC on the 20% BLIND DATA SET SAMPLES SEPARATED AT SOURCE AND NOT PART OF TRAINING 99.5%; 98.5%, 89.5%; 0.98 were the 4 values. Improve: Target Variable by 20%. In this case Not Matched % decrease from 95% to 76% We got decision trees and path for + and -ve outcomes for both current and target outcome values. We got top 5 features with weights for current and target outcomes The results clearly show that our SXI++ algorithm produces precise and accurate results for highly scientific R&D data.

I recently listened to two very different discussions that made me think about the return on investment for scientific instruments that generate ever larger data sets. In one, I listened to discussions about new kinds of mass spectrometers that could generate really large amounts of high resolution data that could be useful in differentiating bacterial strains. The proteomics research that's involved with that is really interesting. In an entirely unrelated discussion held amongst research data managers, they discussed data retention issues of long term data sets. Imagine you have a research activity where instruments generate enormous amounts of data and it is decided that it should be retained long term. Perhaps funds covers the initial generation and storage of data but it doesn't cover the cost of data storage for it a decade later. Where will the money for that come from? Will we reach a point where data storage and maintenance costs get to a point where papers done a decade ago will become deliberately non-reproduceable in order to rationalise long term archival costs? All of this made me reflect on two things that are important in data management. The first comes down to what data sources are 'good enough' for whatever your existing use cases are. When does collecting more data present more of a cost or analytical burden than a benefit? The second is about when it makes sense to invest in a future edge - either in a new line of use cases that require a new minimum dimension of data or uses cases that could re-use a data set beyond its initial purpose. I've seen some good examples of those too but the answer for this isn't simple. #datareuse #datamanagement #proteomics