D ONE – Data Driven Value Creation

Life sciences (Alphabet company):
- Architected and built a DWH on GCP integrating external genic knowledge bases
- Developed a RESTful API to interact with the DWH
- Orchestrated jobs for scheduled data ingestion and modelling
- Maintained and implemented new features for multiple cloud-based products
- Enabled large-scale statistical analysis of gene expression profiles
- Cloudified complex data processing pipelines for large transcriptomics datasets with GKE
- Engineered a cloud-based gene pathway representation build system
- Managed and maintained cloud infrastructure as code with templates
- Performed code review and shared best data engineering practices
- Contributed to multiple cloud-based products

Legal-tech (Swiss startup):
- Performed data-driven analytics on two legal cases
- Increased the number of unit tests for key NLP tasks by an order of magnitude
- Increased F1-score for the named entity recognition task by 18%
- Implemented new approaches to the key NLP tasks, such as NER, entity linking and relation extraction
- Prototyped and then developed a new feature, the knowledge graph

Other domains (construction, aviation, consulting):
- Prototyped a staffing web app implementing retrieval augmented generation (RAG)
- Prototyped a GenAI solution on AWS for transportation regulation
- Advised on building an AI assistant on Azure to facilitate construction site assessment

In addition, I organised and led several workshops on Data & AI topics:
- “Building Agile Data Foundations for Data Mesh & Generative AI”,  IEEE Swiss Conference on Data Science 2024
- “Real World Applications of LLMs for Business and Industry”,  IEEE Swiss Conference on Data Science 2024
- “Optimizing Resource Allocation with LLM-based Solutions”,  Growing Together Conference 2024

Senior Consultant

ITMO University

- Supervised 15+ research projects in cheminformatics and AI for materials science
- Co-authored 2 papers for A* conferences (ICML 2024, NeurIPS 2024)
- Co-authored 5 papers for A* conference workshops (AI2ASE, AI4Research)
- Co-authored 5 papers in domain Q1 journals
- Co-authored 1 paper in a Nature Index journal
- Served as a reviewer for NeurIPS, ICLR, ICML, domain Q1 journals
- Established agile methodology for scientific project management
- Established internal and external collaborations
- Contributed to a series of educational courses

Lead Researcher

ETH Zürich

- Developed RALPS (regularized adversarial learning preserving similarity of samples) for multi-batch mass-spectrometry data normalization. It was tested in three application scenarios on two datasets: a newly generated benchmarking dataset and a real clinical one. RALPS outperformed SOTA methods (EigenMS, WaveICA, NormAE, etc.) in all scenarios. A series of ablation experiments has shown the method is quite robust. Published in Bioinformatics.

- Investigated how different AI paradigms compare in learning representations of biological images. I implemented 4 models sharing the same CNN backbone (weakly-supervised, unsupervised, regularized and self-supervised). Models were trained on a big biological dataset under identical conditions. Learned representations were compared in three downstream analysis tasks. Presented at MIDL 2022 (poster) and published in PMLR.

- Systematically investigated utility of pretrained on ImageNet models (ResNet-50, SwAV, DINO, etc.) in analysing a biological dataset of 1.1M drug-treated cell line images. I evaluated models’ performance with multiple metrics on three tasks and compared to the vanilla convolutional autoencoder newly trained on the entire dataset. I made my model publicly available, since it delivered the best overall scores. I further proposed to use distance-based analysis of learned representations to find temporal patterns common for multiple drugs. Presented at MIDL 2022 (oral) and published in PMLR.

- Designed and implemented a monitoring system to keep track of Agilent 6550 iFunnel Q-TOF mass-spectrometer performance (Python + MySQL + R Shiny). The system is composed of three blocks: i) raw signal processing and feature engineering, ii) statistical data analysis, iii) visualisation and reporting through a web-server. It allowed fast and explicit control of key mass-spec properties, as well as potential applications in mass-spec diagnostics and tuning strategies. Published in Frontiers in Molecular Biosciences.

Scientific Assistant

BIOCAD

- Headed up a new systems biology squad , trained the team of 3 associates and built workflows to deliver analysis of pharmacokinetics (PK) and pharmacodynamics (PD) within limited timelines (e.g., PK/PD modelling results, optimal design of clinical trials).

- Established interactions among the group and the other units: departments of preclinical trials, clinical trials and medicine . We integrated PK/PD modelling into a generic product development pipeline, formalised needs and expectations, fixed terms and timelines.

- Managed and supervised PK/PD modeling for various therapeutics of BIOCAD in different development stages. Within 7 months, we successfully delivered reports on modelling of 10 product projects. Those supported decision making and greatly helped interacting with local and European regulators.

- Managed and participated in the development of our internal computational tool for PK/PD modelling (Python). A few cases of PK/PD modelling were solved using this tool.

Systems Biology Team-Lead

- Designed and implemented approaches for modelling of pharmacokinetics and pharmacodynamics (PKPD) of several therapeutics reaching early stages of clinical trials (Python). Based on modelling, I was able to propose dosage regimens ensuring maximal safety and efficacy. These results helped decision making on the projects and were used in reports for local and European regulators.

- Developed a deep learning model to classify 4 breast cancer stages using microscopy images of immunohistochemistry experiments (Python + Tensorflow). The model achieved 95% CV accuracy. 

- Developed a statistical tool for immunogenicity assessment of antibodies (Python backend). I researched and implemented state-of-the-art methods, and also developed my own method that outperformed the others in precision and robustness (on the available data). I wrote documentation and maintained the tool, supporting about 10 researchers from 3 labs who were actively using it. 

- Developed approaches to increase precision of Gibbs energy calculation using OPLS-type force field (Python). I calculated small molecule parameters that were further used in several in-house computational tools (e.g., for protein folding and docking).

Bioinformatics Specialist

Peter the Great St.Petersburg Polytechnic University

Developed a computational tool for agent-based stochastic modelling of gene regulation processes in Drosophila (Java). I implemented several specific mechanisms, such as short-range repression, and simulated dynamics of transcription factors in the vicinity of gap genes. Statistical analysis of multiple simulations and the scientific conclusions were presented on local conferences.

Research Engineer

Applied mathematician by training with a Master in bioinformatics. Doctor of sciences ETH Zurich with work experience in academy, industry and consulting. Presenter and/or speaker at NeurIPS, ICML, MIDL, AMLD, SDS, BioTechX, and other venues. Skilled in data science and engineering, cloud and software engineering. Experienced in research, project and team management. Passionate about AI and its real-world applications. Willing to set ambitious, yet realistic goals and achieve them together.

Taplio

Andrei Dmitrenko's Linkedin Analytics

Want to drive more opportunities from LinkedIn?

Taplio

.css-1jgreyo{background-image:linear-gradient(284deg, rgba(39,43,72,1) 0%, rgba(33,47,183,1) 100%);color:transparent;-webkit-background-clip:text;background-clip:text;}Andrei Dmitrenko's Linkedin Analytics

Want to drive more opportunities from LinkedIn?

Andrei Dmitrenko's Linkedin Analytics